Accelerating pharmaceutical research with a multi-agent AI system

2026-03-24 · Source: Thoughtworks Insights · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

A multi-agent AI system was developed for a pharmaceutical client to accelerate research by making decades of fragmented data discoverable and usable. The project evolved through three stages: "Search" using semantic search and knowledge graphs, "Ask" incorporating sophisticated retrieval and a conversational UI, and "Do" which utilized multiple agents for research, evaluation, and content creation. Key technical innovations included a four-agent system featuring a critical "reflection agent" to ensure accuracy and comprehensiveness, alongside robust AI evaluation using Langfuse and Ragas for metrics like answer relevancy and faithfulness. Data quality was also enhanced through a process involving PDF extraction, chunking, embeddings, GPT-4 for named entity recognition, and LLM-as-judge evaluation. This system reduced manual searching by 90% and regulatory report drafting from weeks to minutes, avoiding redundant experiments.

Key takeaway

For AI Engineers or Research Scientists tasked with integrating AI into data-rich, regulated environments like pharmaceuticals, consider a multi-agent architecture. Your approach should prioritize iterative development from search to agentic "do" capabilities, incorporating reflection agents for accuracy. Implement robust LLM evaluation using platforms like Langfuse and Ragas, and invest in structuring fragmented data with tools like GPT-4 for named entity recognition to ensure reliable, high-impact outcomes.

Key insights

Multi-agent AI systems can transform fragmented data into actionable insights, significantly accelerating complex research processes.

Principles

• Iterative AI development (Search -> Ask -> Do) improves capability.
• Reflection agents enhance AI accuracy and comprehensiveness.
• Ground truth curation is vital for robust LLM evaluation.

Method

Implement a multi-stage AI system: semantic search, conversational Q&A, and agentic task execution. Enhance data quality via extraction, chunking, embedding, NER, and LLM-as-judge evaluation.

In practice

• Use a reflection agent to self-correct AI outputs.
• Integrate Langfuse and Ragas for LLM evaluation metrics.
• Structure unstructured data with GPT-4 for NER.

Topics

• Multi-agent AI Systems
• Pharmaceutical Research
• LLM Evaluation
• Knowledge Graphs
• Named Entity Recognition
• Data Quality Improvement

Best for: AI Engineer, Research Scientist, Director of AI/ML

Related on AIssential

• AiChemy: Next-Generation Agent with MCP, Skills and Custom Data for Drug Discovery — Databricks enables building custom multi-agent systems by integrating public and proprietary data sources.
• How to Build a Multi-Agent Research Assistant in Python — Multi-agent systems can efficiently produce source-grounded research reports by dynamically escalating research depth based on evidence quality.
• Karpathy Open-Sourced a 24/7 AI Research Lab — Autonomous AI agents can significantly accelerate ML research by independently conducting and optimizing experiments.
• Meta AI - AI at Meta — Advancements span LLM reasoning, vision models, and material science, driven by new benchmarks, datasets, and training methods.
• Towards Generalist Agents for Accelerating Scientific Discovery171 — LLMs can accelerate scientific discovery by generating and refining hypotheses within agentic, iterative workflows.
• Meet an AI Using Scholar — AI tools show promise for systematic review screening and descriptive data extraction, but struggle with comprehensive search and nuanced data interpretation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Thoughtworks Insights.