PyDxAI Development Roadmap / Pathway
Pathway of PyDxAI—its origin, evolution, current status and next steps.
Origin and Foundation
- PyDxAI emerges from the earlier system MIKAI, described as an “in‑house medical AI system built for precision reasoning, document understanding, and clinical support”.
- PyDxAI stands for “Python + Diagnostics + AI” and is positioned as the next generation: “faster, safer, and more modular”.
- The mission: make intelligent, explainable, trustworthy AI a daily companion for doctors, researchers, learners.
Core Pillars / Features
The website outlines major upcoming development pillars for PyDxAI:
- An enhanced inference engine for reasoning across text, images, and structured data.
- A self‐updating knowledge base via curated medical RAG pipelines.
- Fine‑tuned diagnostic dialogue models, especially optimized for internal medicine and endocrinology (which aligns with your specialty).
- A secure integration layer for local hospitals and research networks (important for deployment and data privacy)
Case Study / Diagnostic “Crucible”
- The article describes a real‐world case used as a stress‑test: a patient with evolving symptoms (headache/body aches/cough → travel history → vesiculobullous skin lesion → Tzanck smear) and PyDxAI’s responses at each stage.
- It highlights how PyDxAI retrieved relevant differential diagnoses, incorporated contextual data (travel history), and then refined diagnosis when new definitive lab results arrived.
- It also emphasises a current limitation: ranking/prioritization of likely diagnoses is still imperfect (retrieval works, but clinical weighting needs improvement).
Current Status
- As of October 5, 2025: The website states that PyDxAI is now launched (or announced) and entering a phase of broader research, open collaboration, scalable deployment.
- There is also a Medium blog (by Dr Kijakarn) describing that PyDxAI has achieved a “fully functional Agentic RAG workflow” with intelligent search intent detection, retrieval, integration of live web search, memory update. Medium
Next Steps / Roadmap
From the blog and website, the stated next steps include:
- Fixing JSON encoding / memory insertion bugs (in the blog) for the memory system. Medium
- Enhancing confidence scoring between local vs web‑sourced data. Medium
- Adding summarisation weighting (peer‐reviewed documents must get higher priority). Medium
- Integrating direct API retrieval from platforms like PubMed for evidence‑based references. Medium
- Enabling agentic self‑evaluation, where the system critiques/improves its own answers based on retrieved context. Medium
- On the website side: rolling out multimodal processing (text + image + structured data) and deployment into local hospital/research networks.
Mapping into Architecture (Triple Memory + FrontLLM + PromptBook)
Since we design a system with a triple memory + front LLM (dual LLM) + promptbook architecture, here’s how PyDxAI’s pathway aligns and provides lessons:
- The memory update/self‑learning pipeline of PyDxAI is directly relevant: use local session memory, global medical knowledge, plus web retrieval memory.
- The emphasis on retrieval + ranking + reasoning: PyDxAI’s gap is prioritisation. We should emphasise probability weighting for common vs rare diagnoses (we’ve mentioned this) and integrate it into ranking.
- Multimodal processing: If the system will ingest images (e.g., scanned documents, labs, imaging), make sure image + structured data pipelines are planned — PyDxAI states this as next phase.
- Explainability: PyDxAI emphasises structured audit logs, reasoning transparency — something that should build into UX/reporting module.
- Deployment/integration: PyDxAI’s path includes secure local hospital network integration — this system similarly needs secure architecture, regulatory compliance (especially for medical AI in Thailand).