Skip to content
Rajdeep Mondal logo

About

About Me

I build AI agents that cut operational costs in half.

Right now I am using them to compress drug discovery timelines from years to quarters at Elucidata, where I lead GenAI for six of the top 20 global pharmaceutical companies.

What I Do

I step in when the problem is "we have tried everything."

At Elucidata (~$23M raised, Series A led by Eight Roads), I ship production AI systems under pressure across data, LLMs, and biomedical R&D. The constraint: these systems have to work for drug discovery decisions, not just demos.

Current focus: Ultra Deep Research, an autonomous agent system that reads scientific papers, extracts findings, and synthesizes across sources to cut preclinical research from years to quarters.

Recent results:

  • Built a multi-agent curation engine: ~98% field-level accuracy, 500x faster than manual
  • Co-authored research on multi-agent systems for biomedical metadata at scale
  • Won 1st place ($20k) in NCI's AI Data Readiness Challenge at 95% accuracy
  • Released 23,650 cancer patient phenopackets with NCI and Lawrence Berkeley Lab

What I Build

AI Agents - Systems that reason, plan, and act without hand-holding. I build agents for research, analysis, and workflows where the decision logic has to be explicit, testable, and auditable. The hard part is not getting them to work. It is getting them to fail gracefully.

RAG Systems - Retrieval that works under pressure. Not just vector search: systems that handle bad queries, missing context, and the edge case that breaks everything at 2am. I have built these at scale for biomedical literature search.

Multi-Agent Orchestration - Coordinating multiple agents to solve complex tasks. The hard part is not the agents. It is the orchestration: knowing when to delegate, when to verify, and when to fail.

Production LLM Infrastructure - Taking AI from prototype to production. Evaluation, monitoring, fallbacks, and the unglamorous systems engineering that makes it reliable.

Advising

I advise founders and engineering teams on building production-grade AI agents.

Most agent tutorials teach you to chain API calls. I teach you to build systems that survive contact with real users and real data.

I can help with:

  • Agent architecture for production, not demos
  • Multi-agent coordination and failure modes
  • RAG systems that work when queries are messy
  • Going from prototype to production without burning months

Proof that I know what I am talking about:

  • Systems serving six of the top 20 global pharma companies
  • 98% accuracy on complex extraction tasks at scale
  • Published research on multi-agent systems
  • NCI competition win proving technical depth

If you have tried everything and it still does not work, book a call.

Tools I Use

Languages: Python, TypeScript, SQL

AI/ML: LangChain, OpenAI, Anthropic, Hugging Face, PyTorch

Infrastructure: PostgreSQL, Redis, Pinecone, AWS, Docker

Practices: Agile, CI/CD, test-driven development

Beyond Work

Reading - Psychology, philosophy, startups, business strategy. Currently: Maxwell Maltz, Naval Ravikant, and anything on how great companies get built.

Outdoors - Four Himalayan treks and counting. Goechala is still my favorite. Watching the sun rise over Kanchenjunga is the kind of thing that recalibrates your sense of scale.

Open Source - PMCGrab: a tool that converts PubMed Central papers to structured JSON. Built it because I needed it for biomedical literature parsing at scale.

Work Together

If you are building AI agents and want them to survive production, let's talk.

I am most useful when:

About This Site

I share what I learn: essays on clear thinking, notes from books, technical deep dives on AI engineering, and thoughts on building systems that work.

I write to understand, not to impress. If a piece does not make you smarter, it is not doing its job.