Healthcare2024Confidential client

AI patient intake and triage assistant

A conversational intake assistant that collects symptoms and history before a visit, then hands clinicians a structured summary instead of a stack of free-text forms.

LLM Solutions
Agentic AI

Start a similar project All work

Client: Confidential client
Industry: Healthcare
Year: 2024
Services: LLM Solutions · Agentic AI

The challenge

Clinicians started visits by reading long free-text intake forms.

Front-desk staff were re-typing the same questions every morning: symptoms, duration, medication, allergies. Paper and PDF forms came back half-complete, and clinicians spent the first minutes of each appointment reconstructing a history the patient had already given once.

The clinic wanted patients to answer in their own words, in English or Urdu, without the assistant ever giving medical advice.

What we built

A conversational assistant that gathers symptoms and history, then hands over a structured summary.

A guided chat intake
It runs on web and WhatsApp. The model asks follow-up questions only from an approved question bank, so it can clarify ("how long has the cough lasted?") but never diagnose.
Structured output for clinicians
Each conversation becomes a one-page summary: chief complaint, timeline, current medication, allergies and red-flag answers, with the patient's original words linked beside every field.
Hard safety rails
Red-flag answers such as chest pain or breathing difficulty end the chat and show emergency guidance immediately. The model is not involved in that decision; it is a rule.
Privacy by design
Identifiers are separated from the clinical text before anything reaches the model, conversations are encrypted at rest, and retention follows the clinic's own records policy.

How we tested it

Clinicians wrote the evaluation set: realistic intake conversations, including vague, contradictory and off-topic answers. Every release had to keep the summaries faithful to what the patient said and trigger every red-flag case before it could ship.

Outcome

Clinicians now open each appointment with a structured summary instead of a blank form, and the front desk stopped re-entering intake data by hand. The clinic owns the question bank and updates it without a code change.