For a while I was a web developer who used AI APIs. Then I built a retrieval system for a client's documents, watched it confidently invent an answer, and spent two weeks making it refuse instead. Somewhere in those two weeks the job changed. This is the story of that change, for students who are where I was.
What I was building before
Client sites, landing pages, a CRM for our own agency. React, Next.js, Node, PostgreSQL, Supabase. Good work, real client work — that's where SCULT INDIA got its start. But the interesting problems were on the edges: what the product should do when the data was messy, the user was vague, the answer uncertain.
The project that made it real
The first serious RAG system. Chunking that split tables in half. Vector search that missed exact product names. A model that answered fluently from irrelevant chunks. Each fix — structure-aware chunks, hybrid retrieval, a reranker, a confidence gate — was a problem with no tutorial and a measurable result. That system became RAG.NextUpgrad, and the habit of measuring became the job.
What stayed the same
Everything about shipping. FastAPI and Next.js, Docker, logs, deploys, users who do unexpected things. The web skills were not a detour; they are why the AI systems reached production instead of staying in notebooks. Students who skip the web layer to "do AI" build demos. Students who have it build products.
Why I stayed
- The problems are unsolved. Nobody has the final answer on evaluation, hallucination or cost; there is room to contribute from India.
- Every project teaches something a client will pay for next month.
- The field rewards showing your work — which is how an engineering student ends up speaking at inductions and evaluating other people's hackathon projects.
If you are a web developer wondering whether to make the same move: build one retrieval system for real documents and make it honest. You will know within two weeks whether this is your work. It was mine.
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU Kanpur) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. Email pranjulrathour41@gmail.com.
