# Dhanush Kumar

> I turn legacy software into AI-native systems — in production, not demos. Full-stack engineer in Paris re-architecting a legacy Life Cycle Assessment platform in climate tech; I write weekly about RAG, agents, evals, and the failures in between.

Useful context: eight years of traditional full-stack work (ETL at Deloitte, fintech in India, travel SaaS in Paris) before going AI-native. The writing documents a real re-architecture with the constraints left in — chunking failures, eval sets that gate merges, cost budgets. No course, no newsletter funnel. Contact: dhanushk2105@gmail.com — GitHub: https://github.com/dhanushkumarsuresh-dev — LinkedIn: https://www.linkedin.com/in/dhanushk2105

## Writing

- [Chunking real company docs: what broke and why](https://example.dev/writing/chunking-real-company-docs): Heading-aware chunking works beautifully on documentation written by people who use headings. Real Notion pages are not that.
- [10 questions I asked before building our internal knowledge agent](https://example.dev/writing/ten-questions-before-building-a-knowledge-agent): Before writing any code for our internal RAG wiki, I spent a week asking questions. The questions turned out to be the engineering.
- [Why I'm rebuilding my career AI-native — in public](https://example.dev/writing/rebuilding-my-career-ai-native): CRUD skills became a commodity while I wasn't looking. My answer is to re-architect the platform I work on, and my own career, at the same time — and to document all of it.

Selected posts are also available in French:

- [Chunker de vrais documents d'entreprise : ce qui a cassé, et pourquoi](https://example.dev/writing/chunker-de-vrais-documents-d-entreprise)
- [10 questions que j'ai posées avant de construire notre agent de connaissances interne](https://example.dev/writing/dix-questions-avant-de-construire-un-agent-de-connaissances)
- [Pourquoi je reconstruis ma carrière AI-native — en public](https://example.dev/writing/reconstruire-ma-carriere-ai-native)

## Projects

- [knowledge-agent](https://example.dev/projects/knowledge-agent): Permission-aware RAG over company knowledge: Notion and Drive ingestion, hybrid retrieval, source-linked answers, and an eval set that gates every pipeline change.
- rag-eval-notes: Retrieval benchmarks and chunking experiments on French-language corpora, published as reproducible notebooks.
- this site: Deliberately over-engineered blog: static-first ISR pages with a hand-rolled, Clerk-protected markdown CMS. Source on GitHub.

## Pages

- [About](https://example.dev/about): Career story and what AI-native means here
- [Now](https://example.dev/now): What I'm focused on right now
- [Uses](https://example.dev/uses): Tools and setup
- [Talks](https://example.dev/talks): Talks and appearances
- [Books](https://example.dev/books): Bookshelf with one-line takes
- [AI](https://example.dev/ai): The prose version of this file