Medikle

medikle.com

Medikle is a healthcare platform for medication management. Pruning Labs built its deterministic dosage engine, multilingual patient assistant with clinical audit trails, and the exception dashboard supervising doctors work from - on Next.js 15, FastAPI, Claude, and PostgreSQL with pgvector.

The problem

Complex dosage charts and fragmented portals created constant patient drop-off. Intake ran to five screens, the language was clinical, and anything unusual generated a support ticket.

The constraint that shaped everything: in medication guidance, a plausible-sounding wrong answer is worse than no answer. Nothing generative could be allowed anywhere near a dose calculation.

What we pruned

• Five-step intake pruned to one conversational interaction
• Medical jargon pruned in favour of instant plain-language guidance
• Manual scheduling overhead pruned to zero

What we engineered

• Deterministic dosage engine with zero-hallucination bounds
• Multilingual patient assistant with clinical audit trails
• High-speed exception dashboard for supervising doctors

How it is built

• Dosage arithmetic lives in ordinary, unit-tested code. The model never computes a dose - it explains one the engine has already calculated.
• Patient-facing answers are grounded in an approved content corpus indexed in pgvector, with citations retained on every response for later review.
• Anything below a confidence threshold, or outside the approved corpus, routes to the clinician exception queue instead of being answered.
• Every interaction is written to an append-only audit trail: inputs, retrieved sources, engine output, and what the patient was ultimately shown.

Stack

• Next.js 15
• FastAPI
• Claude 3.5
• pgvector
• PostgreSQL

Disciplines involved