Our Software Engineer Synthia Hunter Achieng recently built a workflow that tackles a common challenge for organizations processing high volumes of applications: verifying that healthcare facilities listed by applicants actually exist and are legitimate.
Watch the full walkthrough: https://www.youtube.com/watch?v=aJciicMaHYM
The Challenge
An organization providing technical assistance to physicians worldwide receives thousands of monthly applications. Each applicant lists their affiliated healthcare site, which staff had to manually verify across multiple countries - a slow, repetitive process that was hard to scale.
The Solution
We combined OpenFn workflow automation with ChatGPT’s Deep Research model to create an AI-powered research assistant for the review team.
Here’s how the workflow works:
-
Reads applicant data directly from Google Sheets (ID, name, site, country)
-
Sends each record to the LLM with a structured prompt asking it to confirm the site exists, identify its online presence, and evaluate credibility
-
Receives a structured JSON response with findings, status, and confidence level
-
Automatically updates the Google Sheet with verification results:
-
Pre-Approved (legitimate site found with high confidence)
-
Declined (no evidence of legitimate facility)
-
Needs Review (insufficient or conflicting evidence)
-
Key Learnings
AI is powerful, but guidance matters Carefully designing the prompt and JSON schema made the difference between vague answers and reliable, parseable responses. We used explicit field names and enforced strict JSON formats.
Human-in-the-loop remains essential The AI triages applications quickly, but ambiguous or low-confidence cases still require human review.
Transparency builds trust Including the model’s reasoning and reference links gave reviewers a clear audit trail for every decision.
Technical Approach
By structuring the AI output as valid JSON, the workflow can programmatically parse results and feed them back into the review system with no manual intervention. OpenFn essentially becomes a bridge between structured applicant data and an AI-powered reasoning engine capable of doing lightweight due diligence on healthcare facilities worldwide.
The blog post includes a full video walkthrough showing how the workflow connects Google Sheets to ChatGPT’s Deep Research model, processes applicant data, and automatically updates verification results.
Read the full post here: AI Workflow Automation for Healthcare Site Verification | OpenFn | OpenFn
Discussion
Have you experimented with combining LLMs and workflow automation for verification or triage workflows? What challenges or learnings have you encountered with prompt design and structured AI outputs?
Join our Wednesday drop-in sessions to discuss AI-augmented workflows: https://addcal.io/e/seijlix5d9a9