Read first, build after !

The first half of 2025 was intense. I dedicated six months to building various LLM-based projects because I believed in the mantra: “build to learn.” Influencers on LinkedIn kept pushing this narrative: “You can’t become an ML engineer without building a full RAG pipeline,” or, “If you’ve never built X, you’re not ready.” Each post would inevitably lead to a pitch for their courses, books, or bootcamps. Honestly, I think this needs to stop.

Still, I learned a lot by building:

• paper_to_podcast: Transforms any research paper into a three-person podcast using Langchain and the OpenAI API.
• brainstormers: A collaborative brainstorming app powered by LLMs (Langchain under the hood).
• aiva: A mock-interview platform that uses your CV to generate real-time, critical feedback.
• dagster knowledge extraction: A data-engineering pipeline extracting structured JSON from text sources via Dagster + any LLM provider.
• LLM story game: An interactive story game where all events are generated dynamically by an LLM.

Despite completing these projects, I realized I had mastered more tools than concepts. LangChain, Langraph, GitHub …etc. I knew them well. But treating everything as a black box felt superficial. Building without theory is not engineering; it’s just clever copying and pasting.

Reflecting again on those LinkedIn influencers, I noticed a troubling pattern: they prioritized tools and trends over fundamental knowledge. This approach made me subconsciously skip reading foundational papers, and I want to end this cycle now.

My New Plan

1. Read Thoroughly I’ll use this excellent repo of classic NLP papers: 100 NLP Papers.

2. Document and Share For every paper I read, I’ll write a clear, straightforward summary post here.

3. Build With Insight After gaining solid theoretical insights, I’ll thoughtfully build practical projects.

4. Balance Theory and Practice My father always said: “Reading and documentation are key. Acquire more data, gain more knowledge, and clarity will replace confusion.”

Conclusion

I will start reading papers thoroughly, acquiring NLP knowledge daily, and documenting my learning in posts on this website. After gaining confidence—perhaps after 1000 papers, though I don’t think that’s enough—I might begin critically reviewing papers. Until then, I’ll simply learn, read, share, and grow. Building blindly without theory isn’t enough; true innovation requires genuine understanding.

As my father always emphasized, documentation and reading are essential. The more data and knowledge you acquire, the clearer your understanding becomes, dissolving the fog of ignorance, confusion, and dissonance. Reading is indeed the key.