Diagnostic AI for Blood & Pathology

I am building an AI-driven application that supports medical diagnostics by reading raw blood-test and pathology reports, extracting the relevant figures, and instantly returning an easy-to-read interpretation along with evidence-based clinical suggestions.

The core workflow is straightforward: a clinician (or patient) uploads a PDF, image, HL7/FHIR payload, or even a snapped photo of a report; the system recognises the data, checks each finding against reference ranges, flags abnormalities, and explains the potential implications in plain language. Where results form recognisable patterns (for example, anaemia profiles or inflammatory markers) the model should surface likely differential diagnoses and recommended next-step investigations.

To get there I will need robust OCR or direct-feed parsing, a well-trained ML/NLP pipeline (TensorFlow, PyTorch or similar), and a lightweight cross-platform front-end—web first, with mobile to follow. Clinical accuracy, privacy (HIPAA / GDPR), and clear audit trails for every prediction are non-negotiable.

Please send me a detailed project proposal covering architecture, data-set strategy, validation metrics, regulatory safeguards, timeline, and your past experience with comparable medical or regulated AI products.

Acceptance criteria:
• ≥95 % field-extraction accuracy on a blind test set of mixed-format reports
• Explanation texts capped at 8th-grade reading level, with source citations
• End-to-end latency under 5 seconds for a typical two-page report
• Complete hand-off package: source code, model weights, and deployment scripts ready for my AWS account

Looking forward to reviewing your approach and seeing how you will turn these requirements into a production-ready diagnostic assistant.