.png&w=256&q=75)
🤓 Latest Submissions
WeCare — Ask Your Medical Records, Safely
The Problem: Patients receive multiple hospital documents (discharge summaries, consult notes) that often contradict each other regarding doses or buried allergies. This is a massive safety risk. Other tools translate a single document. WeCare starts where they stop: making the entire patient record safe to interrogate. Our Solution: WeCare WeCare is a multi-document patient Q&A engine that cross-references all records to prevent medical errors before they happen. Core Features: Deterministic Cross-Document Conflict Engine: LLMs alone aren't safe enough for medicine. WeCare pairs AI extraction with a Python rules engine that scans across all records. It automatically flags dose mismatches (Metformin 500mg vs 1g), buried allergies, and high-risk interactions (Warfarin + NSAIDs). Citation-Bound Q&A: Patients ask natural language questions. The AI refuses to answer unless it grounds the response in a verbatim, line-level citation from the patient's actual records, eliminating hallucinations and allowing immediate source verification. Confidence Gating & Escalation: If a question hits a conflict, or confidence drops below 0.60, WeCare refuses to guess. It routes the query to a Clinician Queue with the AI draft and source lines, ensuring a human-in-the-loop handoff for high-risk decisions. AMD & Google Integration: The pipeline is running live on Google's Gemma-4-31B model, hosted on AMD infrastructure via the Token Factory. The architecture is endpoint-agnostic and optimized for ROCm. Safety First: Synthetic data only. Decision support, not diagnosis. Visible uncertainty metrics. Business Value: A white-label patient portal for hospitals. It reduces triage nurse call volume and directly prevents medication errors. Roadmap: Native EHR integration (HL7/FHIR) for automatic ingestion, location-based emergency routing, and real-time doctor chat.
18 Oct 2026
