Grounded first, model second
Facilities and travel times come from Google's APIs — the model never invents geography; it interprets it.
The Field Guide · room 11 of 11
Med-O-SFT
Finding medical help nearby, with a fine-tuned model in the loop.
Med-O-SFT is a notebook-built medical finder: it chains Google's location APIs to find nearby medical facilities and their travel times, then puts a fine-tuned medical model in the loop to add insights — all wrapped in a Gradio interface.
Built collaboratively with Nishaant Singh; the repository lives under owner Saurabh Kumar's account.
the chain — grounded APIs first, model second
geolocationplacesdistance matrixmedical-o1-sft insights
Facilities and travel times come from real APIs — the model interprets the grounded data; it never invents the geography.
collaborator plate — built with Nishaant Singh · repository owner Saurabh Kumar
The actual problem
Finding medical help nearby is two different questions wearing one coat: where are the facilities and how long does reaching them take — a structured geolocation problem — and what should you know about them — a domain-knowledge question that a generic assistant answers unreliably.
The tool's premise is to keep those questions separate and grounded: real APIs answer the where and how long; the model's role is to add medical insight on top of that verified scaffold, not to invent the geography.
The exploded machine
The pipeline starts with the Google Geolocation API to establish position, queries the Places API for nearby medical facilities, and uses the Distance Matrix API to compute real travel times to each one.
Locator — Google Geolocation API
Position is established from the network — the chain starts where the user actually is.Map roll — Google Places API
Nearby medical facilities are discovered against that position — real places, not guessed ones.Distance drum — Google Distance Matrix API
Travel times to each facility are computed, so “nearby” includes the cost of getting there.Insight lantern — the fine-tuned model call
A fine-tuned medical model (medical-o1-sft) adds insights on top of the grounded facility data.Gradio console — one operable surface
The whole chain is presented in a Gradio interface built up from the notebook pipeline.
A fine-tuned medical model (medical-o1-sft) is called on top of that grounded context to provide medical insights, and the Gradio interface presents the whole chain — location, facilities, travel times, insights — in one console.
Engineering decisions
A three-API chain
Geolocation for position, Places for discovery, Distance Matrix for travel times — each API answers the one question it is designed for.
A notebook-built tool with a real interface
The tool was built as a Jupyter notebook pipeline and surfaced through Gradio — the console makes the chain operable without reading the notebook.
Proof — verified results
- Location chain
- Google Geolocation → Places → Distance Matrixrepo
- Model in the loop
- fine-tuned medical-o1-sft for medical insightsrepo
- Interface
- Gradiorepo
- Clients
- OpenAI / GenerativeAI clientsrepo
- Collaboration
- verified via contributors API, commit history and READMEverified live
every number here carries its source — my resume's figures and the synthetic-fixture results are marked for what they are