The problem
The business development team received medical job postings in wildly inconsistent formats and typed each one into the platform by hand. Every delay in publishing a role lengthened the time a hospital waited to fill it.
- Job details arrived as PDFs, PowerPoint slides, spreadsheets and photos of paper
- Manual transcription was slow and error-prone
- Medical listings needed high accuracy and a person's sign-off before publishing
What we built
An ingestion pipeline that accepts any of those formats and uses a fine-tuned language model to extract the role into a strict schema. Records are created in the platform automatically and queued for a person to review before they go live.
- One upload flow for PDFs, presentations, spreadsheets and images
- A fine-tuned model with custom prompts, constrained to a strict JSON schema
- Records created automatically in the existing platform and queued for human review
Outcome
Manual data entry and review time fell by 76%. Listings are more consistent, roles publish sooner, and the team spends its time on partner relationships instead of retyping.
- Python
- Pandas
- OpenAI API (fine-tuned)
- JSON Schema
- Ruby on Rails