AI Engineer / Agentic Systems / Thomson Reuters
I build production AI agents that do real work — and save real hours.
Agentic systems, RAG infrastructure, and LLMOps for enterprise scale. My agents have eliminated 50,000+ hours of manual work a year and influenced $1M in revenue — measured, evaluated, and running in production.
Discipline dial: drag or use arrow keys to point at agents, RAG, LLMOps, or evals.
Drag the dial — it responds
Hours of manual work automated annually
Revenue influenced by the asset generation engine
Manual effort cut by the autonomous interviewing agent
Concurrent users served by the retrieval infrastructure
Section 02 / Practice
I’m Anubhav Santra — an AI engineer who builds agents that hold up in production. Not demos: systems with evaluation harnesses, hallucination controls, and the plumbing to serve thousands of users at once.
At Thomson Reuters and ZS Associates I’ve shipped autonomous interviewing agents, multi-agent generation engines, and document-intelligence layers — measured in hours saved, revenue moved, and error rates down: 50,000+ hours automated annually, $1M revenue influenced, errors cut by up to 50%.
I work across the full stack of applied AI: LangGraph and MCP orchestration, hybrid RAG, LLMOps CI/CD, and the eval frameworks that make model behaviour a number instead of a feeling.
The best agent is the one your team stops noticing — the work just happens.
What I build
Autonomous agents, multi-agent orchestration (LangGraph / MCP), hybrid RAG pipelines, and LLM evaluation harnesses that run in CI.
How I work
Measure first, then build. Every system ships with evals — faithfulness, precision, recall — and feedback loops wired for production.
Currently
AI Engineer at Thomson Reuters, building enterprise agentic systems.
Open to consulting & selective opportunities
A small archive of identities, interfaces, and digital systems built to make an idea easier to understand — and harder to forget. Each specimen below is the live work, not a screenshot.
trace output — idle
Agentic AI / Enterprise automation
An agent that interviews your documents: parses meeting transcripts, asks for what’s missing, auto-fills confirmed data, and ships the swimlane report. 90% of the manual effort, gone.
Output · email / clinical
— idle · press Generate —
Eval framework · live
Grounded by hybrid retrieval.
Claims mapped to source tables.
Hallucination gates in CI.
Multi-agent generation / RAG
A multi-agent system that writes pharma marketing assets end-to-end — agent handoffs, tool orchestration, and hybrid RAG over OpenSearch. 66% less manual effort, $1M revenue influenced.
Document outline · click a section
query: ITT primary endpoint at 12 months · hybrid retrieval
retrieving…
3/3 chunks above relevance floor · document p.214 of 1,024 · ~5 min/doc full-parse
RAG infrastructure / MCP tools
The knowledge layer for enterprise agents — MCP retriever tools that turn 1,000+ page multimodal scientific documents into section-level answers in minutes, scaled to 10,000+ concurrent users.
Every project begins with a question, a constraint, and a reason to exist. This is the path so far.
Building production agentic systems on the enterprise knowledge stack — autonomous interviewing agents, MCP tooling, and self-improving agent harnesses.
Shipped multi-agent systems for US pharma: asset generation, brand analytics on Databricks, and a document knowledge layer scaled to 10,000+ concurrent users.
Graph neural networks for clinical data — from research notebooks to production NLP/ML pipelines.
ML-accelerated phase-field simulations — encoder–decoder forecasting with ConvLSTM on 10k+ images.
Medical imaging — CNN classification pipelines for skin-cancer datasets, above state of the art.
Recognition tied to shipped systems — every certificate is one click away.
Section 05 / Contact
Have a workflow that eats hours?
Selected consulting, product collaborations, and full-time AI engineering roles — all routes open.
santralabs.xyz
iamanubhav2001@gmail.com