NLP & Generative AI Engineer
Work on Kalpana Chapter and KisanMitra: retrieval, grounding, and language that meets people where they actually are.
- Ref
- J04
- Team
- Platform
- Where
- Remote, anywhere in India
All roles
Kalpana Chapter talks to students in crisis, in Hinglish, at three in the morning. KisanMitra answers a spoken question about a diseased crop in the language it was asked in. Both are retrieval problems before they are generation problems, and both fail badly if the model invents an answer.
This seat is about grounding: making a system that says "I do not know" instead of something fluent and wrong.
What you would do
- Build and tune retrieval over real source material, not toy corpora
- Reduce hallucination by construction — grounding, citation, refusal paths
- Evaluate output quality in languages you may not personally speak
- Handle Indic and code-mixed text as a first-class case, not an afterthought
What we are looking for
- Python, and hands-on work with LLMs and embeddings
- You have built retrieval that had to work, and seen where it breaks
- Understanding of why a confident wrong answer is the dangerous failure
- Bonus: you speak an Indian language other than English
These are not calculator projects or website clones. You will work on RAG systems, real databases and production AI pipelines — the kind of work you can actually talk about in an interview.