Anukriti

Sentinel Clinical is an autonomous AI investigation engine for adverse drug events. When a patient suffers an adverse drug reaction, severe neutropenia after fluorouracil therapy for example, Sentinel doesn't answer a question. It runs a multi step clinical investigation: plans what evidence to gather, retrieves relevant documents via VultronRetriever (ranked #1 on the ViDoRe V3 visual document retrieval benchmark), calls deterministic pharmacogenomics tools (anukriti pgx core), generates competing hypotheses with honest confidence scores, drafts a structured report, and subjects it to a skeptical AI Reviewer before finalizing. If the Reviewer rejects, say because a CPIC guideline is population level evidence not patient specific proof, it triggers a re investigation loop. If evidence is missing entirely, Sentinel refuses to conclude rather than guessing. The core design principle: the LLM explains, deterministic rules decide. Built on LangGraph multi agent orchestration, Vultr Serverless Inference (VultronRetriever for multi round evidence retrieval, Kimi K2.6 for reasoning), Azure AI services (Document Intelligence, AI Search, Cosmos DB), and deployed on a Vultr VM.