AI Case Review Platform Developer

We're a Michigan personal injury law firm with five offices and hundreds of active files. We're building our own AI-assisted case review platform, and we'll own the code, prompts and data. You'll build it on a commercial AI model through its API, with OpenAI to start. We are not asking anyone to build or train a model. Phase 1 is a 90-day pilot. It takes a complete case file and produces, for attorney review: • A case summary in our template • A medical and case chronology with a page citation for every entry • A list of missing records, bills, insurance and lien information, checked against our checklist • A priority level and a recommended action list for the case manager and attorney You'll also build OCR for scanned and handwritten records, plus a review queue where attorneys approve, edit or reject each output. Our staff need an editor to change checklists and templates without code. The build also needs a test harness that scores outputs against our answer keys, and per-file cost logging. Optional: a management dashboard built on our case-management data. Must-haves: • You've shipped AI applications that work on long, document-heavy files: OCR, search across long files, answers with citations. Show us. • At least two people on your side will know the code. • Everything is built in accounts we own: cloud, code repository, AI provider. Code goes into our repository from day one. • Full IP assignment to us, with disclosure of any existing or open-source code you use. • Security fit for medical records: encryption, single sign-on with multi-factor login, role-based access, audit logs, and development on de-identified files only. • Payment by milestone. Each payment is released after our independent technical reviewer signs off on its acceptance tests. Nice to have: • Legal or healthcare clients • Experience with Business Associate Agreements and zero-data-retention AI setups • Integrations with law-firm case-management systems How we'll choose: shortlisted applicants get our full RFP. Up to three finalists complete a paid test project: a medical chronology with page citations, built from a de-identified record set and scored against an answer key. How to apply: start your proposal with the word "chronology" so we know you read this. Answer the screening questions and attach examples of relevant work. We won't review generic proposals. Screening questions 1. Describe one AI application you shipped that processes long documents. What did it do, how many pages did it handle, and how did it show its sources? 2. How would you make sure every entry in a medical chronology cites the right page, and keep unsupported statements out? 3. How do you handle scanned and handwritten medical records? 4. Who are the two people who will know this code, and what are their roles? 5. Do you accept full IP assignment and working only in client-owned accounts? If not, what are your exceptions?