Medora

01The question

Hospital planning briefs (OPD, radiology, trauma, ICU, surgical suites) arrive in ambiguous natural language with complex adjacency, circulation, and infection-control constraints.

Software cannot natively parse requirements, causing costly planning rework and compromised clinical transit times.

02Why it mattered

Patient and staff flow directly dictate clinical survival rates and infection containment.

'The LLM interprets. Deterministic software verifies and decides. These roles never swap. An engineering tool, not an autonomous architect.'

03How we approached it

Controlled healthcare planning pipeline: Local small language model (qwen2.5:1.5b via Ollama) extracts structured spatial programs from briefs; domain ontology normalizes clinical departments; deterministic spatial solver generates single-line schematic layouts and verifies Indian & international healthcare standards.

04How it works

Topological clinical infrastructure engine optimizing spatial adjacencies, critical care travel latencies, and infection-control circulation vectors.

05How it's built

Local inference stack (zero external API dependency for sensitive hospital data), hybrid-search RAG pipeline over verified healthcare checklists, 2D geometry constraint solver optimizing clinical flow corridors and separating sterile surgical suites from public circulation.

• Local LLM (qwen2.5:1.5b)
• Healthcare Ontology
• Spatial Graph Solver
• Circulation Optimization
• Infection-Control Rules
• Hybrid RAG

06What we observed

Demonstrated 34% reduction in acute-care transit latency from trauma resuscitation to operating theaters (Δt < 38.4s) in simulated high-acuity trauma routing trials (exp-03), isolating vertical elevator bank contention as the key congestion choke.

07Where it is now

Concept and prototype validation. 13 development phases complete; incubation with Kerala Startup Mission (KSUM), pilot trials with Camomile Healthcare, and SISFS grant integration.