PharmaGuide

Inspiration

Every 8 minutes a patient is harmed by a drug interaction a pharmacist was too overloaded to catch. Pharmacists juggle hundreds of prescriptions a shift; interactions hide in combinations no human can reliably hold in memory. We wanted to give them a second set of eyes that never blinks.

What it does

PharmaGuide is an AI pharmacist co-pilot. It screens any prescription against a clinical-grade interaction knowledge base in real time, flags conflicts, explains the mechanism behind each one, and suggests safer alternatives a pharmacist can approve in one tap. It turns the pharmacy counter into an early-warning system without slowing anyone down

How we built it

A FastAPI backend serving a structured interaction engine, with an LLM layer for patient-friendly explanations and alternative suggestions, grounded in the clinical drug database. Prescription parsing handles drug names, doses and combinations; the engine resolves interactions and severity, then the LLM drafts the 'why' and the swap options, constrained to the knowledge base so it never hallucinates a recommendation. The frontend is a clean Next.js interface built for speed at a busy counter.

Accomplishments we're proud of

A system where AI is genuinely central, not bolted on: detection is deterministic and auditable, explanations are generative. The one-tap approval flow keeps the pharmacist in charge, which is how it gets adopted.

What we learned

In healthcare, the AI's job is to be fast and honest, never the final authority. Grounding beats autonomy, and the best interface is the one that respects the human expert.

What's next for PharmaGuide

Real-time insurance formulary checks, patient-specific flagging (age, pregnancy, kidney function), and a pharmacy-network pilot to measure prevented adverse events at scale

AI Tools Disclosure for PharmaGuide

We used AI/ML at every layer of the stack:

• Interaction engine: structured clinical drug-interaction database with deterministic resolution logic (no hallucination risk on the core check)
• LLM layer: a large language model (OpenAI API) for parsing messy prescription inputs (brand names, abbreviations, dose formats) and generating patient-friendly interaction explanations and safer-alternative suggestions, constrained to facts verified by the structured engine
• Frontend acceleration: Google Stitch
• Development workflow: AI-assisted debugging and code review throughout The core safety decision, whether a drug pair interacts and how severe it is, is deterministic and auditable. AI is central to parsing and explanation, but never the final authority: every recommendation requires pharmacist approval

Built With

• claude
• fastapi
• python
• stitch