Freelance AI Engineer — RAG & GenAI Specialist ( shouldn't be engaged in full time roles)

Overview

*Contract / Freelance India (On-site) 3–6 month engagementImmediate startUS client · Legal domain | 40 Hours / Week *

** Candidates shouldn't be engaged in full time roles **

About CellStrat

CellStrat builds production-grade AI solutions for healthcare, banking, and legal sectors. Our Healthcare AI Platform, CellAssist, powers clinic and hospital operations end-to-end — from appointment booking through billing. We partner with US and global clients to deliver bespoke AI systems that move beyond prototypes and into real operations.

The engagement

A US-based legal client has an existing RAG pipeline for legal document processing and needs an experienced engineer to expand it — adding new rules, agents, and custom features. You will join an active AI team with full ownership of key components and ship production-ready code from day.

This is not a greenfield build. You are inheriting an existing codebase and extending it thoughtfully. Strong code-reading skills and the ability to ramp up fast on an existing system are essential.

What you will work on

Extend and harden the existing legal document RAG pipeline — chunking strategies, retrieval tuning, re-ranking, and source citation improvements
Design and implement new LLM agents for specific legal document workflows (clause extraction, contract comparison, compliance checks)
Integrate additional rule engines and custom features as defined by client requirements
Build and expose FastAPI endpoints to support downstream client applications
Collaborate with the CellStrat AI team and US client stakeholders to translate requirements into implementation plans (HLD/LLD)
Write production-quality code with proper testing, documentation, and deployment via CI/CD on cloud infrastructure
Monitor and optimise pipeline performance — latency, accuracy, token cost
Must have
Strong Python engineering — clean, testable, production-ready code. FastAPI for API development.
Hands-on RAG experience — not tutorial-level. You have built retrieval pipelines, chosen chunking strategies, tuned semantic search, and evaluated retrieval quality in production.
Vector store experience — Milvus, Pinecone, Weaviate, FAISS, ChromaDB, or similar. Know when and why to choose one over another.
Working knowledge of LLM frameworks — LangChain, LangGraph, LlamaIndex, or similar. Ability to design agentic workflows, tool-calling, and multi-step reasoning pipelines.
System design capability — ability to produce HLD/LLD documentation and translate business requirements into architecture decisions independently.
Track record of shipping production-quality AI solutions — not just PoCs. You can point to live systems you have owned end-to-end.
Cloud deployment experience — AWS, Azure, or GCP. Docker, CI/CD, basic infrastructure ownership.

Tech stack (expected familiarity)
PythonFastAPILangChain / LangGraphLlamaIndexOpenAI / Anthropic APIsVector DBsPostgreSQLDockerAWS / Azure / GCPCI/CDGit

Good to have

Prior experience in legal tech, document processing, or contract intelligence (clause extraction, NER on legal text, compliance automation)
Knowledge graph or ontology-based retrieval (Neo4j, entity linking, GraphRAG)
Experience with retrieval evaluation frameworks — RAGAS, LLM-as-Judge, or custom eval pipelines
Prompt engineering discipline — structured prompts, output enforcement, hallucination mitigation
Bachelor's or Master's degree in Computer Science, Engineering, or a related field

Engagement details
Type
Freelance / Contract
Duration
3–6 months (extendable)
Location
On-site, India
Start
Immediate
Client
US legal tech client
Seniority
Senior / Mid-senior only

We are not looking for entry-level talent. This role requires someone who understands the AI landscape, can architect thoughtfully, and ships without hand-holding.

To apply, reach out with a brief note (3–5 sentences) describing your most relevant RAG or GenAI project — what you built, what stack you used, and what you shipped. Resumes without this note will not be reviewed.