An agentic AI platform for drug discovery
Specialized scientific agents on Microsoft Discovery Studio that mine literature, validate targets and screen molecules — with a scientist approving every step and every tool call on an immutable audit trail.
- Role
- Agentic AI Technical Architect
- Period
- 2026 – present
- Stack
- Literature, target validation, screening
- Multi-agent
- Agent tool calls on an immutable audit log
- 100%
- Controls enforced in the platform
- GxP · HIPAA · FDA
Context
Pharmaceutical R&D teams spend a large share of their time on work that is necessary but slow: reading literature, cross-referencing target pathways, and preparing candidate molecules for screening. The goal was an agentic platform that would speed up hypothesis testing for global scientific teams — without ever taking the scientist out of the decision.
The platform was co-engineered with Microsoft product teams on Microsoft Discovery Studio.
Constraints
- Regulated environment. GxP, HIPAA and FDA expectations apply to how data is accessed and how decisions are recorded.
- Heterogeneous data. Assay results, target pathways, chemical and molecular structures, and genomics — structured and unstructured — all needed to be retrievable.
- Scientists stay accountable. Agents may propose; people decide.
Architecture
- Specialized scientific agents running on Azure AI Foundry Hosted Agents, using Claude via the Anthropic API for reasoning and synthesis.
- Multi-agent orchestration with action-based tool calling and custom tool-integration frameworks, automating literature mining and synthesis, target validation and molecular screening.
- A data layer built for retrieval: secure REST APIs and connectors bringing R&D data into Azure Data Lake, Discovery Studio storage and GraphRAG knowledge stores, so agents can reason over relationships — not just text similarity.
- Human-in-the-loop checkpoints wherever an agent’s output would change what scientists do next.
Guardrails & governance
- Prompt safety filters and execution guardrails on what each agent may do.
- Strict RBAC, so agents only reach the data their task requires.
- Immutable audit-trail logging of every agent tool call — the answer to “why did it do that?” is always on record.
My role
Strategic technical lead and primary technical point of contact for Microsoft product engineering and the client’s R&D, Bio-IT and Data Science leadership — aligning platform capabilities with pharmaceutical research roadmaps.
Client identity withheld. Details are limited to what can be shared publicly.