R&D Case Study — Agentic AI for Drug Safety
SafeSignal AI — Human-Governed Pharmacovigilance Automation
A multi-agent pharmacovigilance workflow that transforms unstructured adverse-event information into traceable, reviewer-ready safety cases with duplicate detection and E2B(R3)-aligned output.
R&D prototype evaluated on synthetic safety reports. This is not a production medical device, clinical decision system, or autonomous regulatory-submission tool.
The Challenge
Drug-safety reports arrive through fragmented channels and require careful validity checks, seriousness triage, duplicate detection, terminology mapping, and evidence-backed narratives. General-purpose automation can accelerate the work but may omit critical facts or invent missing information, making traceability and human control essential.
What We Built
- Specialist LangGraph agents for intake, validity checks, clinical extraction, seriousness triage, duplicate detection, narrative drafting, and quality review
- Field-level provenance linking every extracted value and narrative statement to its original evidence
- Hybrid controls combining language models with deterministic validation, confidence thresholds, and mandatory human approval
- Temporal case graph and E2B(R3)-aligned structured output for reviewer-ready safety operations
LangGraphFastAPIPostgreSQLpgvectorNeo4jn8nOCRWhisperReactDockerOpenTelemetry
Synthetic Prototype Benchmark
Serious-event detection recall
96.2% on 180 held-out synthetic cases
First-pass preparation time
18 min manual → 3.4 min assisted
Reviewer correction rate
11.8% across 1,260 extracted fields
Results are illustrative figures from a controlled synthetic-data prototype evaluation. They are not independently validated production results and do not represent performance in a live pharmacovigilance operation. Human approval remains required before export or downstream action.
More Work
Batch Disclosure — Price-Moving Catalyst Detection
An AI-powered system that scans ASX and SEC filings for forward-looking disclosures — contracts, M&A, guidance — that move micro-cap prices before the market catches on.
WikiGen — Autonomous, Bias-Aware Knowledge Synthesis
A multi-agent GraphRAG system that autonomously writes reliable, bias-aware Wikipedia-style articles with verifiable provenance on every claim.
Let's Work Together
Explore a Safer AI Workflow?
Start with a bounded, synthetic-data pilot and a jointly defined validation plan.