From Product Manager to Open Source Author: My Journey of Technology Entrepreneurship

The journey of technology entrepreneurship of the elderly product manager Dugu Shrimp. From technical background to product transformation in 2003, from amateur development open source project Lorn.ADSP to AI era reconstruction, from midlife crisis to technical author transformation. Share how AI tools such as DeepSeek can help leap development efficiency, the creation process and market response of the two technical monographs, and provide transformation ideas and practical references for middle-aged technical people.

Introduction: A product manager’s technical complex

I am Dugu Shrimp, a product manager, I am in my forties this year, and I have a girl who is in elementary school at home. Although I am now a product manager, I have always been a programmer at heart – after graduating in 2003, I worked in development and architecture for 6 years, and only switched to product in 2009.

It’s a bit embarrassing to say, after being a product manager for so many years, I still can’t change my love of researching technology. Every time I see the code of my development colleagues, I can’t help but want to point fingers, and sometimes I even ask directly: “Why is this interface designed this way?” Can you optimize it? “I guess many development colleagues feel that I, as a product manager, is a bit “out of line”. But there is no way, the technical complex is like being engraved in DNA, and it can’t be erased.

This “restless” technical complex eventually led me to do something crazy – I started an open source project called Lorn.ADSP, a .NET-based advertising platform. Looking back now, this road has been quite tossed, but the gains are also real. Sometimes when I wake up in the middle of the night, I still think: Fortunately, I didn’t listen to my wife’s advice back then, and I honestly became a stable product manager, otherwise how could I tell so many stories now?

Part 1: Original intention and confusion – why do this open source project?

The entanglement of technological transformation

Speaking of open source projects, we have to start in 2013. At that time, I had just jumped from Youku to another Internet video company, responsible for the construction of advertising system products from 0 to 1. Speaking of the experience on Youku, it’s quite interesting to think about it now.

Our leaders are ruthless people from technical backgrounds, and the requirements for product managers are particularly high – not only to understand the business, but also to understand the code. Every once in a while, I would hold a technical debriefing meeting, let me talk about the architecture and algorithm ideas of the project in front of a bunch of R&D, and finally ask R&D to help see if I understood correctly. At first, I was really nervous, for fear of being laughed at for saying the wrong thing. But there is no way, you have to bite the bullet.

In order not to lose face in front of the tech bosses, I gnawed the source code of their advertising system from beginning to end. During that time, I had meetings to discuss requirements, and at night I went home to study the code, sometimes it was midnight. My family says I’m crazy, I’m obviously a product manager, why do I have to toss myself like this?

As a result, I found a lot of problems! What frustrated me the most was that the advertising strategy system was too rigidly written, basically a bunch of if-else hardcode. I thought, how can this continue? If you want to add a new strategy, isn’t it necessary to re-publish it? It’s too much trouble. And those conditional judgments are written in a mess, and it takes a lot of effort for newcomers to understand.

B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …

View details >

The ideal architecture in my mind should look like this: abstract the advertising strategy into interfaces and base classes, implement different subclasses for different strategies, and then use dependency injection to make the whole system flexible and configurable. This makes adding strategies as easy as building blocks, and the code structure is much clearer.

However, I am a product manager after all, and my words are not so important. And to be honest, I haven’t actually verified whether this idea works.

The desire to practice true knowledge

Another reason why I want to do it is that although I have been working on advertising system products all these years, I always feel that there is a layer of paper. Product managers look at problems more from a business perspective, and the details of technical implementation are often reported by R&D colleagues, lacking that kind of deep understanding.

I participated in the upgrade of keyword strategy when I was at Baidu, and I led the restructuring of the advertising system in Suning. Just like learning to swim, just watching others swim is completely different from going into the water yourself. Sometimes I suddenly think of a technical problem in the middle of the night, but I can’t verify it immediately, which is really uncomfortable. It’s like you know the answer is there, but you just can’t reach it, which is very crazy.

There is also a more realistic consideration. After making products for so many years, I find that there are really few people who can understand business and technology at the same time. Most product managers know little about technology, and drawing a prototype is enough; Most technical students are not sensitive to business logic, and the functions they implement are always not right. And I have 6 years of development experience and many years of product experience, why not try to combine these two samples to do something?

The establishment of the original intention of open source

Based on these ideas, I decided to do a project myself, with a simple goal:

Verify that my idea of the ad delivery system architecture is feasible (mainly to prove that I am not just talking and practicing)
Truly understand the technical nature of the advertising system through a complete practical process
Provide a relatively reliable reference implementation for other developers
By the way, I will exercise my full-stack ability from product to technology

It’s easier said than done, I know it’s a big pit, but the stubbornness of the technicians made me decide to jump down and try. Anyway, the evening and weekend time are also idle, so why not use it to do something interesting? The worst result is nothing more than not being able to do it, but at least you can learn a lot in the process.

Part 2: Technology selection and architecture design – building an intelligent advertising platform

Selection of core technology stacks

In terms of technical selection, I chose it without hesitation. .NET technology stack, mainly developed in C# and F#. Speaking of which, there is a bit of personal feelings in it – when I first graduated in 2003, I used it. .NET started with ASP.NET 1.1 all the way to .NETFramework, and has a natural intimacy for this technology stack. Although Java was already quite popular at that time, I still felt that . .NET is easier to use.

A friend asked me why I didn’t choose Java, after all, most Internet companies use Java. To be honest, it’s not that I have any prejudice against Java, it’s purely because . .NET I’m more familiar with. Doing open source projects is originally done in spare time, and if you have to spend time relearning the technology stack, the progress will be too slow. And to be honest,. .NET is really not bad in terms of performance and development efficiency.

Of course, choose. .NET is not only a feeling, but also a practical consideration:

The performance is really powerful: Anyone who has done advertising systems knows that millisecond-level response requirements are no joke. The goal I set for myself is to complete the request processing within 50ms, supporting 100,000+ QPS, . .NET has been doing well in this regard. Especially later. After NETCore came out, the performance improvement was more obvious.
High development efficiency: When a person engages in a project, efficiency is the lifeline. VisualStudio’s smart prompts, rich class library, and perfect debugging tools all allow me to avoid a lot of detours. Sometimes a function can be done in a few minutes with the shortcut keys of the IDE. If it is replaced with other technology stacks, the environment configuration alone is enough for me to toss for half a day.
Cloud deployment friendly: Although it is a personal project, I designed it according to production-grade standards from the beginning. .NET support for containerization and microservices makes deployment relatively simple. Later, the rise of Azure also gave .NET more advantages in cloud services.

System architecture design concept

Lorn.ADSP uses a microservices layered architecture, which is a best practice that I have summarized over many years. The entire system strictly follows IAB industry standards and supports OpenRTB real-time bidding protocol.

Several core principles of architecture design:

The performance must be ruthless: The response time of the ad request must be less than 50ms, and it supports 100,000+QPS, which is the bottom line and is not negotiable. When I was at Baidu, a technology expert once said to me: “If the advertising system is one millisecond slower, the company will earn 10,000 yuan less.” “It’s a bit exaggerated, but it’s right.
Availability should be high: The system availability is at least 99.9%, and it supports automatic recovery of faults, which is responsible for the user and for himself. I have seen too many loss of advertising revenue due to system failures, and I understand the feeling of distress that has been made in advertising products.
The standards should be strict: Strictly follow the IABOpenRTB protocol, support VAST/VMAP video advertising standards, and have all the anti-cheat mechanisms. This is not for show, but to ensure the system’s versatility and scalability.

To be honest, it’s really not easy for a person to do all of this in place, but it’s interesting because of the challenges, isn’t it?

Core functional module design

After much thought, I split the system into several core modules:

Advertising business management: Customer relations, contract management, financial settlement, these basic functions, although not sexy, are essential. This part is relatively simple, mainly CRUD operations, but there are many details. For example, the effective time of the contract, the billing method, the settlement cycle, etc., each field has business logic in it.
Statistical analysis: This is the innovation point I want to verify the most. Traditional advertising systems are rule-driven, and I would like to try to use statistical analysis algorithms to predict what kind of products and advertisements users prefer, thereby increasing click-through rates and conversion rates. Including user behavior analysis, product recommendation algorithms, delivery effect prediction, etc., these are quite novel ideas in 2013.
Ad serving engine: Real-time bidding, intelligent decision-making, multi-objective optimization, which is the heart of the whole system. It’s simple to say, but it’s really complicated to do. There are several algorithms for bidding strategies alone, and various constraints such as budget control, frequency limits, and blacklists must be considered.
Data analysis BI: Multi-dimensional analysis, predictive insights, custom reports, and data-driven are the soul of the advertising system. I refer to some design ideas from Google Analytics and Baidu Statistics, but it is much simpler.

Looking back now, the design concept at that time was quite forward-looking, but the technical conditions limited the realization of many ideas. Some functions were considered important at the time, but when they actually did it, they found that there were too many pitfalls, so they had to be put on hold first.

Part 3: Development Journey and Challenges – A Technical Marathon for One

Initial technical challenges

After the project started, I realized that I had kept things simple. A person has to be a product manager, an architect, and a programmer, and he switches roles frequently like juggling, and sometimes he really feels like he is about to become schizophrenia.

Demand sorting stage: I have to put on the hat of a product manager and sort out the needs of the advertising system from scratch. This process is quite interesting, I have studied mainstream platforms such as Google Ads and Facebook Ads and analyzed their functional features. Sometimes I suddenly think of a good idea in the middle of the night, quickly get up and write it down in my notebook, and when I get up the next day to look at it, some ideas are really good, and some are just cranky.
System design phase: Then change to architect mode and translate business requirements into technical solutions. How to split microservices? How to design a database? What is the API API specification? These questions come up one after another, and each of them needs to be carefully considered. Sometimes for the design of a table structure, I can sit in front of the computer all afternoon, draw, change, and toss repeatedly.
Coding implementation phase: Finally, it’s time to code. To be honest, this stage is the most painful and fulfilling. The pain is because time is limited, and the progress is always too slow; The sense of accomplishment is because it is really cool to see lines of code turned into a working system. Especially when I saw the system running for the first time, I was so excited, just like when I wrote “HelloWorld” for the first time.

Breakthrough in technical difficulties

There were many more pitfalls than expected during development:

Performance optimizationThe most headache. The advertising system has extremely strict requirements for response time, and 100ms is already considered slow. In order to achieve the goal, I put a lot of effort into the caching strategy, designed a multi-level caching system, and did a lot of data preprocessing optimization. Sometimes I can spend the night thinking about code to optimize the response time of a few milliseconds. The most exaggerated one, I debugged for three nights in a row to figure out why a SQL query was so slow, and finally found out that there was a problem with the index design. The feeling of sudden realization is really cool.
Algorithmic model integrationis another big pit. In 2013, machine learning was not as mature as it is now, and it was really not easy to integrate statistical analysis algorithms into production systems. Finally, I adopted a hybrid strategy: lightweight algorithms for core decision-making and offline algorithms for accuracy for complex analysis. It’s simple to say, but there are all kinds of pitfalls to achieve.
Microservices architecture managementIt’s also a challenge. There are more and more services, and dependencies are becoming more and more complex, so it is necessary to design a reasonable governance mechanism. Service registration discovery, configuration management, link tracing, fuse downgrade, these lofty concepts must be implemented one by one. At that time, the concept of microservices was relatively new, and many best practices were explored by myself.

Exploration of project management

Traditional project management methods are not suitable for personal projects, so I have developed my own method:

time management: You can squeeze out 2-3 hours every night, 6-8 hours on weekends, and about 20 hours a week. Time is limited and must be used efficiently, so I have a very strict plan for myself. Sometimes my family sees me working so hard and thinks I’m going to change jobs. In fact, it is a simple technical fetish.
Version planning: Split the project into multiple versions, each focusing on a specific function. In this way, we can not only ensure continuity, but also see the phased results, so as not to hit the enthusiasm too much. There is a small sense of accomplishment with each completed version, just like beating the game.
Technical debt management: When time is tight, it is inevitable to leave some technical debts, so I have specially built a list for regular review to ensure that the project will not be dragged down by technical debt. This habit will also be useful later at work.

This methodology was later used in my other projects, and the results were not bad. Now that I think about it, doing open source projects is really a good self-management training.

Part 4: Technological Reconstruction in the AI Era – Embracing the Wave of Artificial Intelligence

DeepSeek’s Discovery and Application

In 2024, with the rapid advancement of AI technology, especially the rise of domestic open source large models such as DeepSeek, I see an opportunity to reconstruct Lorn.ADSP. Coincidentally, it happened that I was laid off at that time, and I suddenly had a lot of time to re-examine the project.

Being laid off is quite depressing. I have 20 years of experience in advertising system products, I have worked in major manufacturers such as Baidu and Suning, leading the upgrading and transformation of DAU platforms of more than 100 million, and there is also an open source project called Lorn.ADSP on GitHub (although there are only about 100 stars and about 50 forks, but in the vertical field of advertising system, this number is not bad), in theory, you should not worry about working. But the reality is very skinny, and when I started submitting resumes again, I found a cruel fact: age is indeed a threshold.

However, the loss of the horse in the Saion is not a blessing. Because of the more time, I was able to calm down and rethink the direction of Lorn.ADSP. The emergence of DeepSeek gave me a new inspiration.

The importance of technological sovereignty: As a product manager who has worked in a large factory for many years, I know the importance of data security and technology autonomy and control. DeepSeek’s open source nature and localization background allow us to localize training data and control algorithms independently, avoiding the risk of data export that may be faced by using foreign models. This is significant for advertising systems, which involve large amounts of private user data.
The attractiveness of cost advantages: Compared with the international mainstream business model, DeepSeek can reduce the cost of use by more than 80% at the same level of capability. This is a huge advantage for entrepreneurial projects. I calculated that if I use OpenAI’s API, the training and inference costs alone will be enough for me to buy a new computer.
Recognition of technical capabilities: Through actual testing, I found that DeepSeek excels in code generation, text understanding, multimodal processing, etc., and can fully meet the technical needs of Lorn.ADSP. Especially when it comes to code refactoring, I was pleasantly surprised by DeepSeek’s performance.

AI-driven system refactoring

Based on DeepSeek’s technical capabilities, I started the AI refactoring of Lorn.ADSP:

Intelligent idea generation system：

Multimodal Content Generation: Supports AI automatic generation of text, image, and video ad creatives
Brand Style Adaptation: Automatically adjust creative styles based on brand tone and target audience
Creative effect prediction: Use large models to analyze historical data and predict creative performance
Dynamic Creative Optimization: Analyze user feedback in real-time to automatically optimize creative elements

This part of the feature excites me a lot. In the past, advertising systems relied entirely on manual creativity to generate, not to mention low efficiency, and the quality was uneven. Now with the blessing of AI, high-quality ideas can be generated in batches, and they can be continuously optimized based on user feedback.

Intelligent analysis of user profiles：

Deep user understanding: Construct three-dimensional user portraits through natural language processing and behavior analysis
Intent Recognition Prediction: Predict user purchase intention and interest changes based on user behavior sequences
Personalized recommendations: Provide personalized advertising content and delivery timing for each user
User Lifecycle Management: Intelligently identify user value and lifecycle stages

Intelligent optimization of delivery strategies：

Smart Bidding Strategy: Automatically adjusts bidding strategies based on real-time market data and user value
Intelligent budget allocation: Use large models to predict the delivery effect of different time periods and channels and optimize budget allocation
Anti-cheat detection: Intelligent detection and prevention of ad cheating through abnormal behavior pattern recognition
Performance attribution analysis: Analyze advertising performance in multiple dimensions and provide intelligent attribution analysis

These features were only reminiscent of in 2013, and now they are finally realized. While there are still many details to be perfected, the basic framework is already in place.

A leap in development efficiency

Using DeepSeek for assisted development has significantly improved my development efficiency:

Code generation and optimization:D eepSeek can quickly generate high-quality code frameworks based on my requirements and provide optimization suggestions. This greatly reduces my time investment in writing basic code. Sometimes for a complex feature, I just need to describe the requirements, and DeepSeek can give a good implementation plan. Of course, the code quality still needs to be reviewed manually.
Documents are automatically generated: The system’s technical documentation, API documentation, etc. can be automatically generated through DeepSeek, which not only improves efficiency but also ensures the quality and consistency of documentation. The biggest headache in the past was writing documents, but now it’s much easier.
Test case design:D eepSeek can automatically generate test cases based on code logic, improving the test coverage of the code. While the generated test cases also require manual adjustments, they save a lot of time.
Problem diagnosis and resolution: When encountering technical problems, DeepSeek can quickly analyze the cause of the problem and provide a solution, greatly reducing the debugging time. There were a few particularly tricky bugs, but through conversations with DeepSeek, I quickly found a solution.

Honestly, having an AI assistant feels like having a round-the-clock senior programmer partner who is ready to discuss technical issues. The experience was really great.

Part 5: Knowledge Precipitation and Achievement Transformation – From Project to Monograph

The birth of the first book

During the development of the Lorn.ADSP project, I have accumulated a lot of practical experience in intelligent marketing and AI applications. In particular, deep thinking in advertising, user portraits, recommendation algorithms, etc. gave me the idea of systematically organizing this knowledge.

In fact, the idea of writing a book has been in my mind for a long time. Every time I share some experience in the technical group, or do technical sharing within the company, my colleagues always say: “Your content is quite valuable and you should write a book.” “At first, I didn’t think so, I felt that I was just an ordinary product manager, how could I be qualified to write a book? But as I gained experience, especially with the Lorn.ADSP project, I found myself with some unique insights.

In 2023, I completed my first monograph, “Intelligent Marketing: How Large Models Empower Operations and Product Managers”, published by Tsinghua University Press. Speaking of which, this process is quite interesting, I first wrote an outline and submitted it to the publishing house, but I didn’t expect the editor to be very interested after reading it, and it was quickly approved. But when I really started writing, I realized that it was much more difficult than I imagined to systematically sort out the knowledge in my head.

The planning and creation of the second book

At the end of 2024, with the rapid development of domestic AI models such as DeepSeek, I began to plan my second book, “DeepSeek Application Advanced Tutorial – Product Manager + R&D + Operation + Data Analysis”. The motivation for the creation of this book mainly comes from:

Insights into market demand: Through the survey, it was found that more than 70% of Internet practitioners face the pain point of “insufficient tool familiarity leads to waste of efficiency” and urgently need systematic AI application guidance. This data was obtained by me through questionnaires and interviews, and although the sample size is small, it is very telling.
Trends in technological development: AI technology has evolved from general Q&A to post-specific workflows, and the in-depth application of vertical scenarios has become a new growth point. In my experience with DeepSeek, I have learned that generic prompts are often mediocre, and only those optimized for specific scenarios can exert real value.
Accumulation of personal experience: Combined with my experience in developing the Lorn.ADSP project and writing my first book, I have the ability to provide readers with more practical AI application guides. And this time, with the help of AI assistants, the writing efficiency has also improved a lot.

This book adopts a three-dimensional architecture of “post-task-tool”, covering the AI empowerment solutions of four core positions in the Internet industry: from the intelligent generation of PRD by product managers, competitive product monitoring, to the full-cycle assistance of code for technology development, to the content creation factory and user insight system of operation, and finally the automation engine and risk management brain of data analysis.

Unexpected market reaction

To be honest, the market performance of DeepSeek Advanced Tutorial exceeded my expectations. Just half a month after the book was launched, it rushed to the 35th place in the Dangdang computer/network new book list. When I swiped this ranking that night, I really couldn’t believe it, and I sent the screenshot to my wife, and she was even more excited than me.

What surprised me even more was that this book also entered the CITIC Bookstore Airport Store, which is known for its strict selection. CITIC Bookstore’s selection standards have always been very high, and being able to enter their bookshelves shows that the quality of the content has been professionally recognized. Once I passed by the airport bookstore on a business trip and saw my books in a prominent position, it was a really wonderful feeling, just like seeing my own children perform on stage.

What makes me most gratified is that several universities have ordered this book as a textbook. When I received the first purchase notice, I had mixed feelings. I remember the confusion and uneasiness when I first became a product manager – I obviously have a technical background, but I always feel that something is missing on the road to product. Now that the cross-border experience of technology and products can be written into AI application textbooks recognized by universities, this transformation from practitioner to knowledge disseminator is really emotional.

This feedback reinforces my belief that the market really needs such practical AI application guides. It is not a deep algorithm theory, but a specific method that can be directly applied to work. Seeing readers share cases of using the methods in the book to improve work efficiency in the group, the sense of accomplishment is more real than any business success.

Part 6: Midlife Crisis and Restart: From Setbacks to Growth

An unexpected blow in the workplace

In 2024, just as I was immersed in the sense of accomplishment of technical creation, reality slapped me hard – I was laid off.

This blow is indeed not small. I have 20 years of experience in advertising system products, I have worked in major manufacturers such as Baidu and Suning, leading the upgrading and transformation of DAU platforms of more than 100 million, and there is also an open source project Lorn.ADSP on GitHub (although there are only about 100 Stars and about 50 forks, but in the vertical field of advertising system, this result is actually good), and in theory, you should not worry about working.

But the reality is very skinny. When I started submitting resumes again, I discovered a cruel fact: there are indeed many positions of advertising system product leaders on recruitment software, but my resumes are basically sunk into the sea, and occasional replies are official words such as “inappropriate” and “position mismatch”.

During that time, I was really confused: Is it really because I am older? The annual turnover of the advertising system I have been responsible for is billions or tens of billions, and the ability should not be worse than those small and medium-sized companies, right? Do today’s companies really only look at age and not ability? Sometimes I wake up in the middle of the night and think about these questions, and the more I think about it, the more I can’t sleep. My family was worried when they saw my state. My daughter once asked me, “Dad, are you out of work?” At that moment, my heart was really sour.

Re-examine your worth

After being angry and confused, I began to think calmly. Although the midlife crisis was painful, it also gave me the opportunity to re-examine my worth.

Uniqueness of technical capabilities: Although the market is not friendly enough to middle-aged employees, compound talents like me who can do products and technologies at the same time are still valuable. Especially in the era of AI, this cross-border capability is even more scarce. Most product managers don’t understand technology, most technicians don’t understand products, and I can do both.
Depth of industry experience: 20 years of experience in advertising systems is not in vain, and the understanding of this field is indeed deeper than that of young people. This kind of experience accumulation is precipitated over time, not obtained in a short period of time. I have stepped on every pit, I have walked every detour, these are wealth.
The ability to learn is still there: I was able to quickly master these new technologies of DeepSeek, which shows that my learning ability has not deteriorated due to age. Sometimes I feel like I learn faster than some young people because I have the foundation and experience to support me.
The manifestation of creative ability: The publication of two books proved that I had an advantage in summarizing and disseminating knowledge, which opened up a new career path for me. The ability to write was something I hadn’t realized before, and now I’m discovering that it may be more enduring than technical ability.

New thinking for transformation

Since the traditional job search path does not work, let’s change the way of thinking:

Indie developer: Concentrate on doing a good job in the Lorn.ADSP project and making it a truly valuable open source product. Now that you have time, it’s better to really put this project in place.
Technical author: Continue to deepen in the field of AI applications and provide practical guidance to more practitioners. The success of the first book gave me confidence that this path would work.
Education and training: Develop AI application training courses based on practical experience. Nowadays, enterprises have a great demand for AI training, which is a good direction.
Consulting services: Provide AI transformation consulting for small and medium-sized enterprises. Many companies want to use AI but don’t know how to use it, and my experience comes in handy.

These directions can give full play to my compound advantages and allow me to continue to shine in the field of technology. The layoff incident became an opportunity for me to transform. Sometimes life is like this, it seems bad, but in fact it is the beginning of new opportunities.

Part 7: The Evolution of Full-Stack Product Managers – A Complete Practice from 0 to 1

Redefine the boundaries of product managers

After the midlife crisis, I began to rethink the boundaries of the profession of product manager. Traditional product managers are mainly responsible for demand analysis, product design, project management, etc., and technology implementation mainly relies on the R&D team. But in the age of AI, I’ve discovered a new possibility – full-stack product managers.

The so-called full-stack product manager must not only have the ability of a traditional product manager, but also be able to complete the technical implementation independently. The value of this capability lies in:

Deeper product understandingWhen you can implement the product function yourself, you will have a deeper understanding of the product and be able to find many problems that were overlooked in the design stage. I modified the original product design countless times during the development of Lorn.ADSP, because it was only found out which designs were unreasonable at the time of implementation.
More efficient communication: When communicating with the R&D team, be able to accurately express requirements in technical language to avoid understanding bias. In the past, when I discussed technical solutions with my development colleagues, there was often a situation where chickens and ducks talked, but now it is much smoother.
Faster verificationFor some innovative ideas, prototypes can be quickly built for verification without waiting for R&D resources. This advantage is particularly obvious, sometimes I can make a prototype in half a day with an idea, and if I follow the normal process, it may take a week or two.
Stronger competitiveness: In an entrepreneurial environment with limited resources, full-stack capabilities can greatly enhance individual and team efficiency.

The empowering value of AI tools

During the restructuring of Lorn.ADSP, I deeply experienced the empowering value of AI tools for full-stack development. AI tools like DeepSeek allow me to:

Learn new technologies quickly: Through dialogue with AI, you can quickly understand the core concepts and application methods of new technologies. Sometimes when I encounter a new technical concept, I ask DeepSeek directly, and the explanation it gives is often easier to understand than looking at the documentation.
Code efficiently: AI can generate high-quality code frameworks based on demand descriptions, greatly improving development efficiency. While the generated code still needs to be adjusted, the basic framework is right and saves a lot of time.
Automate document generation: Technical documents, user manuals, etc. can be automatically generated through AI to ensure the quality and timeliness of documents. The biggest headache in the past was writing documents, but now it’s much easier.
Intelligent problem diagnosis: When encountering technical problems, AI can quickly analyze the cause and provide solutions. There were a few particularly tricky bugs, and by describing the phenomenon and pasting code snippets, DeepSeek quickly came up with solutions.

The combination of these capabilities allows me to do everything from requirements analysis to product launch in one person. This was unthinkable before.

Restart Lorn.ADSP

With more time and AI tools, I decided to re-systematically develop the Lorn.ADSP project. This development has clearer goals and better planning:

Technical architecture upgrade: Redesign the system architecture based on the latest technology stack, and make full use of AI capabilities to improve the intelligence level of the system.
The product positioning is clear: No longer a simple technical verification project, but to create an open source advertising platform with real commercial value. This time I will take it seriously and make it as a real product.
Attach importance to community construction: Strengthen interaction with the open source community and attract more developers to participate in project contributions. Although there are only about 100 Stars and about 50 Forks now, this has proven the value of the project, and I will continue to work on this basis.
Business model exploration: Explore the commercialization path of open source projects to achieve sustainable development of projects. Pure open source is difficult to sustain, and a business model must be found to last.

The construction of the knowledge system

In the process of redeveloping the project, I am also building my own body of knowledge. This system includes:

Technical capability matrix：

Back-end development: .NET, microservices, database design
Front-end development: React, Vue, responsive design
AI applications: large model integration, prompt word engineering, model fine-tuning
O&M deployment: Docker, Kubernetes, cloud services

Product design methodology：

User needs analysis
Product function design
Data-driven decision-making
A/B testing implementation

Project management practices：

Agile development process
Version planning management
Quality control system
Risk identification and response

Business thinking cultivation：

Market analysis capabilities
Business model design
Competitive strategy formulation
User value creation

The construction of this knowledge system also provides rich material and profound insights for my two books.

Part 8: The Value and Impact of Knowledge Dissemination – From Personal Growth to Industry Empowerment

From technical practitioners to knowledge disseminators

Looking back on the past few years, I find that I have unknowingly completed an identity transformation: from a pure technical practitioner to a knowledge worker with both practical and dissemination skills. This transformation process has given me a deeper understanding of the value of knowledge dissemination.

The compounding effect of knowledge: When you write an experience into a book and a course, its value will be infinitely magnified. One person’s practical experience can help thousands of people avoid stepping on the pit through knowledge dissemination, and this value amplification effect is unmatched by any other work.
The power of reverse learning: In the process of writing the book and answering readers’ questions, I found that my understanding of many concepts became more profound. In order to explain a technical point to others, you must first thoroughly understand it yourself, and the effect of this “teaching and learning” is very obvious.
The establishment of industry influence: Through continuous knowledge output, I have established a certain influence in the field of AI applications. Nowadays, companies often invite me to share, and the technical community invites me to participate in discussions, which is an influence that money cannot buy.

Inspiration from reader feedback

After the publication of the two books, I received a lot of feedback from readers, which inspired me a lot:

Diversity of needs: The readers in different positions have completely different concerns. Product managers are more concerned about how to improve work efficiency, technology development is more concerned about specific implementation plans, and operation students are more concerned about content creation skills. This made me realize that the application scenarios of AI tools are richer than I imagined.
Commonality of pain points: Although the positions are different, the core pain points faced by everyone are very similar – they all don’t know how to really use AI tools, rather than staying at the level of simple conversations. This also validates the value of writing these two books.
The urgency of learning: Many readers expressed a strong need for learning and hope for more practical cases and operation guidance. This opened my eyes to the huge market potential for online education and corporate training.

Exploration of the commercial value of knowledge dissemination

With the increase of influence, I also began to explore the commercialization path of knowledge dissemination:

The demand for corporate training is strong: Many companies want to equip employees with the use of AI tools, but lack systematic training courses. I started to take on some corporate internal training projects and found that this market is in great demand.
Extension of consulting services: Based on the methodology in the book, I provide AI transformation consulting services to some small and medium-sized enterprises to help them formulate AI application strategies and implementation plans.
The value of community operations: I have established an AI application practice group to regularly share the latest technical trends and application cases. This community not only helps readers solve problems, but also provides me with first-hand information on market needs.

Methodological summary of technical writing

Through the writing practice of the two books, I have summarized a set of technical writing methodologies:

Scenario-driven content organization: Don’t start from the technology itself, but from the actual scenario of the user. Describe the problem first, then introduce the solution, and finally give specific steps.
The value orientation of cases is king: No matter how perfect the theory is, it is not as persuasive as a real and usable case. Each case I provide in the book is real-world validated, ensuring that readers can apply it directly.
Continuously iterating content mechanisms: Technology is developing too fast, and the traditional publishing model cannot keep up with the pace. I have established a content update mechanism to provide readers with the latest supplementary material online.
The value of reader co-creation is amplified: Encourage readers to share their own application cases and experiences, forming a virtuous circle of knowledge co-creation. This not only enriches the content but also enhances the activity of the community.

Thoughts on knowledge work in the AI era

Through the practice of knowledge dissemination in the past few years, I have some in-depth thoughts on knowledge work in the AI era:

The timeliness of knowledge challenges: AI technology is developing too fast, and traditional knowledge dissemination methods can no longer keep up with the pace. We need to establish a more agile knowledge update mechanism.
Combination of practice and theory: The knowledge dissemination value of pure theory is limited and must be combined with practical application scenarios. This requires knowledge disseminators to have both deep theoretical skills and rich practical experience.
The importance of personal branding: In the era of information overload, personal brand has become an important carrier of knowledge dissemination. Readers prefer to trust experts with practical combat experience rather than theorists on paper.
The trend of community-based learning: The traditional one-way knowledge dissemination is changing to community-based learning. Knowledge communicators are no longer high-ranking experts, but organizers and facilitators of learning communities.

Conclusion: The balance between technical feelings and realism

Looking back on this journey from product manager to open source author, the biggest feeling is the art of balancing technical feelings with real life. On the one hand, my love for technology and my pursuit of perfect products have never changed; On the other hand, the midlife crisis and market reality have also made me more grounded.

The Lorn.ADSP project has been on and off since 2013, and although there are only about 100 Stars and about 50 forks on GitHub, which doesn’t seem like much, I know that behind each Star is a developer who really pays attention to this project. It carries my obsession with technology, maybe it will never be the next Spring Boot, but it records a product manager’s understanding and exploration of technology. Honestly, that’s enough.

The publication of these two books has allowed me to find another way to realize value. Being able to share your experience and thinking with more people and help them work better in the AI era is more real than a promotion and salary increase. Especially when I get feedback from readers that my book really helped them, it was a really great feeling, just like the excitement I had when I first saw Lorn.ADSP forked.

In this era, AI technology is developing so fast, and each of us is facing new opportunities and challenges. Age can be a stumbling block to job searching, but knowledge and ability are wealth that will never depreciate. Through continuous learning, practice and sharing, you can always find your place in this era.

For those who are also facing career confusion in middle age, I would like to say: don’t give up your love for technology because of a momentary setback, and don’t doubt your value because of the indifference of the market. Times are changing, but the need for truly capable people will never go away. The key is to maintain a learning mindset, embrace change, and create value for the world in your own way. Sometimes the road seems difficult, but as long as you persevere, there will always be a turnaround.

Technology changes the world, and each of us can be an agent of that change. This is my open source path and my growth experience in the AI era. I hope my story can bring some inspiration and encouragement to more people.