OncoAI – Multi-Agent AI for Precision Cancer Treatment
🧬 I built a multi-agent AI system for oncology on AWS — and it just made the Global Top 1000 in AWS AIdeas 2026.
Here's the architecture:
5 specialized agents run sequentially:
→ Genomics Agent — interprets variants, classifies driver/passenger mutations
→ Oncologist Agent — guideline-aware treatment reasoning
→ Trial-Matching Agent — live clinicaltrials.gov API with eligibility logic
→ Toxicology Agent — adverse effect prediction, drug interaction flags (our differentiator)
→ Compliance Agent — final veto authority, regulatory enforcement, blocks unsafe outputs
AWS stack: Bedrock (LLM explainability) · Lambda · API Gateway · DynamoDB · SQS (agent handoffs) · S3 · CloudFront · CloudWatch
The hardest design decision wasn't the AI — it was making safety override intelligence at every layer. A Confidence Decay Model reduces output confidence by 30% on safety concerns, and 100% on compliance blocks. No unsafe output ever reaches a clinician.
Built to FDA CDS Guidance, HIPAA, and ISO 14971 risk management principles. Explicitly not a diagnostic or prescriptive system. Human oversight is mandatory by design.
Full article here 👇 — would mean a lot if you read it and voted. 🙏
#AWSAIdeas #AWSCommunity #AmazonBedrock #MultiAgentAI #HealthcareAI #ServerlessArchitecture #BuildOnAWS
Replies