AI Home-Loan Eligibility Assistant

I’m building an in-house AI that will instantly tell a customer how much of a home loan they qualify for, list any missing paperwork, and flag likely queries an underwriter would raise—all from the documents the customer uploads plus the data we already store in our customer database.

Here is the scope I need covered:
• Ingest a PDF, image, or text bundle supplied by the applicant, run OCR/NLP to extract key financial and KYC fields, then cross-check those against the records that sit in our existing customer database (SQL).
• Apply a rules-plus-ML approach to calculate loan eligibility in real time for home-loan products only, following the criteria we will share (income ratios, credit behaviour, property value, etc.).
• Produce two outputs: (1) an eligibility summary with sanctioned amount range and reasoning, and (2) a checklist that shows “Provided / Missing” documents together with a “Probable queries” section that mirrors an underwriter’s typical clarifications.
• Surface the results via a lightweight REST/GraphQL API and an internal web dashboard so the credit team can review or override.

Acceptance criteria
– ≥95 % field-extraction accuracy on our validation set.
– Eligibility predictions must match current manual decisions within ±5 % on a three-month back-test.
– Response time under 3 seconds for a standard five-document upload.
– Codebase delivered in Python (TensorFlow / PyTorch acceptable) with clear setup docs, Dockerfile, and unit tests.

I will supply anonymised historical loan files, our decision rules, and database schema. You will handle model development, integration, and knowledge transfer. If you have prior work in document AI or lending automation, that will be a big plus—share live links or repos so I can evaluate fit.