Skip to content

Milestones Status

BB8 Flame Project Roadmap

This roadmap outlines the major milestones, development phases, and ongoing priorities for our clinical graph-based AI platform. It’s designed to give contributors, collaborators, and stakeholders a transparent view into what’s been built, what’s in progress, and what’s coming next. We’ll keep this page up to date as the project evolves—so whether you’re following along or looking to get involved, you’ll always know where things stand.

This phased roadmap will guide development from internal prototypes to robust public releases, with a focus on interpretability, real-world validation, and clinical alignment.

🎯 Goals

Our mission is to empower high‑stakes decision‑making through AI that is patient-centric, accurate, contextually grounded, and transparently reasoned. To fulfill this mission, Diagnostic Agent pursues the following core goals:

  • Embed Agentic Autonomy with Human Oversight
    • Design a modular, multi-agent architecture capable of navigating graph queries, recovering from retrieval failures, and intelligently rerouting reasoning as needed. Importantly, our goal is to institute tiered human supervision, particularly at critical junctures such as treatment recommendations, medication prescriptions, or patient procedures, where licensed clinicians must review and sign off before any action. This approach aligns with best practices in health-care AI governance, where automated suggestions guide physicians but final decisions remain firmly with humans, minimizing automation bias and preserving patient safety through supervisory protocols and asynchronous oversight models.
  • Demonstrate Graph‑Enhanced Accuracy
    • Build a retrieval‑augmented generation pipeline that uses structured knowledge graphs (i.e., GraphRAG) to improve factual grounding, especially for multi-hop reasoning and complex queries, compared to traditional vector‑only RAG systems. Research shows GraphRAG delivers significantly better precision and reasoning coherence by linking entities and relationships rather than isolated text chunks. For AI applications, GraphRAG has also been shown to significantly reduce the hallucination rate, especially in complex domains like healthcare, where factual understanding of the relationships between conditions, symptoms, and treatments is critical.
  • Enable Explainability and Traceability
    • Generate decision traces and reasoning pathways that are human‑interpretable: each answer should be accompanied by the chain of graph queries, agent actions, and retrieved evidence that supports it. This transparency is critical for trust, reduction in missed diagnoses, improved coverage of rare conditions, and auditability.
  • Measure Real‑World Impact
    • Define and track meaningful evaluation metrics—such as retrieval accuracy, diagnostic consistency, model confidence alignment, and clinician trust—to assess how the system performs in actual healthcare settings. These metrics help evaluate downstream impact in contexts like frontline triage, patient case review, clinical supervision workflows, and mobile health deployments in low-resource environments.

Together, these goals reflect IDinsight’s broader commitment to evidence-based innovation. They outline how Diagnostic Agent integrates structured knowledge, agentic reasoning, and rigorous evaluation to create a medical diagnostic assistant that is not only powerful and scalable, but also explainable, trustworthy, and aligned with an impact‑driven mission.

🚦 Current Phase

  • Phase 2: Multi-agent, Benchmarking, and MCP
  • Start Date: 2025-08-04
  • Key Activities:
  • Deploy and integrate Diagnostic Agent with HEP Assist from Last Mile Health
  • Package Diagnostic Agent as an MCP-compatible server, enabling any compliant LLM client or tool to connect
  • Finalize reflection and rerouting logic in multi-agent graph traversal
  • Incorporate SOAP-style notes for physician messaging
  • Evaluate patient messages from Last Mile Health against RAG and GraphRAG pipelines
  • Evaluate synthetic patient messages using GraphRAG pipeline
  • Implement multi-agent pipeline with Pydantic Graph; backend APIs with FastAPI

🚀 Release Plan

Phase Timeline Core Deliverables Release Milestone
Phase 5 – Impact & Continuous Learning Q4 2026+ Impact dashboards, community tools, v1.x iterations Community Roll‑Out & Iterative Releases
Phase 4 – Scale‑Ready Release Q1–Q3 2026 Optimized performance, integration toolkit, governance plan, prep field teams Version 1.0 Clinical-grade Release
Phase 3 – Public Beta Q3-Q4 2025 Public API/UI, onboarding guides, pilot feedback loop, accuracy improvements, documentation Public Beta Launch (v0.9)
Phase 2 – Multi‑Agent, Benchmarking, and MCP Q2–Q3 2025 Multi‑agent orchestration with self‑reflection modules, benchmarking, and MCP servers Beta‑ready Feature Freeze
Phase 1 – Prototype & Internal Release Q2 2025 Graph‑based backend, single‑agent pipeline Internal Alpha Release
Phase 0 – Planning & Ideation Q1 2025 Mission, Use Cases, Architecture Blueprint Internal Design Freeze

đź’¬ Get Involved

Want to contribute or stay in the loop?

We welcome collaborators, testers, and clinical partners to help shape this effort!