The GenAI engineer who ships.
Five production AI applications built end to end in one stretch: a RAG platform, a fine-tuning platform, document intelligence, in-browser vision and a speech workspace. Each one deployed, monitored and used.
- Based
- India · remote, hybrid or Delhi NCR
- Stack
- Python, FastAPI, Next.js, PostgreSQL
- Models
- OpenAI, Claude, Gemini, Mistral, Groq, Llama
- Also
- Campus speaker, hackathon judge
- 5production AI apps
- 14engineers led
- 2+years shipping AI
- 10+hackathons
What I build
The four kinds of system I have taken from idea to production.
- 01
Retrieval-augmented generation
Hybrid dense and BM25 retrieval, reciprocal-rank fusion, re-ranking, streaming answers with citations, and a confidence gate that refuses to guess.
- 02
LLM fine-tuning
QLoRA through Transformers, PEFT and TRL, live training telemetry in the browser, and base-versus-tuned evaluation before anything ships.
- 03
Document and speech AI
Schema-validated extraction from invoices and contracts, page-batched OCR, live speech-to-text and document-scoped chat with page citations.
- 04
Computer vision in the browser
Face detection, recognition and liveness with ONNX Runtime Web; the server keeps embeddings only, never an image.
How I work
The habits behind every one of those systems.
- 01
Evidence before generation
Retrieval quality, schemas and evals come first; the model is the last component I trust, not the first.
- 02
The whole stack
FastAPI and Next.js, PostgreSQL and vector stores, Docker and CI/CD with structured logging and metrics from day one.
- 03
Shipped means operated
Each service ran in production with monitoring, provider fallback and a free-tier footprint, not a notebook.
Proof
Where the claims come from.
- NextUpgrad
- GenAI engineer, Feb to Jul 2026: five AI applications shipped end to end
- SCULT INDIA
- Leading engineering for an AI-first studio and its 14-member team, 2024 to now
- Hackathons
- First prizes at IIT Roorkee (Changethon 2025), CSJMU and MeetKats; top 3 at IIT Kanpur
- Writing
- Guides and articles on RAG, fine-tuning and shipping AI at pranjulrathour.github.io
Read next
Pages that go deeper.
Twenty minutes to see if it's a fit.
Role, team, stack and timelines. Google Meet or phone, Monday to Saturday, India time.