What I'm Doing Now

A now page. Not a resume. A snapshot of where my time actually goes.

The Mission

I am building AI agents to compress drug discovery timelines from years to quarters.

The constraint: preclinical research moves slowly because humans are the bottleneck for reading, synthesizing, and reasoning across biomedical literature. Agents can do this faster and more thoroughly. The question is whether they can do it reliably enough to trust. That is what I am solving.

Day Job

Senior Data Scientist at Elucidata

Elucidata has raised ~$23M, including a$16M Series A led by Eight Roads, with participation from F-Prime, IvyCap, and Hyperplane. Our customers include six of the top 20 global pharmaceutical companies.

We turn raw biomedical assets into AI-ready data by uniting omics, clinical, and imaging sources so R&D moves faster.

My role: I lead GenAI and step in when the problem is "we have tried everything." I ship production systems under pressure across data, LLMs, and biomedical R&D.

What I'm Building

Ultra Deep Research

The headline project. Goal: cut preclinical research timelines from years to quarters.

How it works: autonomous agents that read scientific papers, extract findings, identify contradictions, synthesize across sources, and generate hypotheses. Not summarization. Reasoning at scale.

The hard part is not getting the agents to produce output. It is getting them to produce output you can trust when the stakes are a drug development decision.

Biomedical Data Findability

An LLM-powered search system across open-access PubMed (full text plus supplements), GEO, ArrayExpress, PRIDE, ClinicalTrials.gov, and 8+ other sources.

The problem: biomedical researchers spend more time finding relevant data than analyzing it. We are fixing that.

Multi-Agent Curation Engine

Built an engine that ingests messy biomedical documents, generates metadata schemas, extracts fields, and normalizes to controlled ontologies.

Results:

• ~98% field-level accuracy
• Up to 500x faster throughput than manual curation
• Research paper: "Multi-agent AI System for High-Quality Metadata Curation at Scale"

Previous Systems

Before this, I built the core biomedical knowledge graph, search infrastructure, and open-source fine-tuning infrastructure at Elucidata.

Publications & Research

Multi-agent AI System for High-Quality Metadata Curation at Scale Research paper outlining the system and benchmarks for scalable, high-quality biomedical metadata curation.

Oncopacket: Integration of Cancer Research Data Using GA4GH Phenopackets Co-authored with collaborators at NCI and Lawrence Berkeley National Laboratory. We open-sourced the code and released Phenopackets for 23,650 individuals across 12 cancer types to enable standards-based downstream AI/ML analysis.

Competition Wins

1st Place: NCI CRDC AI Data Readiness Challenge (Multimodal AI/ML)

Won $20,000 with a three-person team. Our model separates primary tumor from normal solid tissue in lung squamous cell carcinoma at ~95% accuracy with strong class-imbalance handling.

Advising

I advise founders and 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.

What I help with:

• Agent architecture for production, not demos
• Multi-agent orchestration and failure modes
• RAG systems that work when queries are messy and context is incomplete
• Taking AI from prototype to production without burning months

Credibility:

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

How to engage:

• Book a 30-minute call to discuss your problem
• If you have tried everything and it still does not work, I am probably the right person

Open Source

PMCGrab - Converts PubMed Central papers to structured JSON. Useful for parsing biomedical literature at scale. Built it because I needed it.

Learning

• LLM fine-tuning - When prompting is not enough. The question: what is the minimum data needed to see real improvement?
• Evaluation frameworks - How to know if an AI system is actually working. Most evals test what is easy to measure, not what matters.
• Drug development - Completed the Drug Development Product Management Specialization (UC San Diego). Understanding the domain I am building for.

Reading

• Psycho-Cybernetics by Maxwell Maltz (finished) - Your self-image is the thermostat. Change it, and behavior follows.
• The Almanack of Naval Ravikant (in progress) - Wealth, happiness, and compound advantage. Dense with insight.
• Build by Tony Fadell (up next) - How products get made by someone who made the iPod.

Writing

• A deep dive on practical RAG patterns: what works, what does not, and why most tutorials skip the hard parts
• Book notes, one per month: the ideas that survived the edit
• Essays on building AI systems that work outside of demos

Location

Bangalore, India. Remote.

Work Together

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

I am most useful when:

• You have a real problem, not a technology shopping list

• You have tried the obvious solutions and they did not work

• You need someone who has shipped systems at this scale before

• Email: rajdeep@rajdeepmondal.com

• LinkedIn: @rajdeep-mondal

• Book a call: Schedule 30 min

Last updated: January 26, 2026