Financial due-diligence consultancy.

The problem

This Nordic consultancy does financial due diligence — and much of what a deal hinges on comes out in interviews. Every engagement works through a structured due-diligence agenda of questions, and each interview produces a long transcript. Between the two sits an analyst, reading every transcript line by line to find where each question on the agenda was answered: slow, expensive work that varies from analyst to analyst, and that can quietly miss the moments that matter most in a deal — the things management volunteers without being asked. And because these are confidential, deal-sensitive interviews, the obvious shortcut of pasting transcripts into a public AI chatbot was never an option.

What I built

A pipeline that reads interviews the way a diligent analyst would — only automatically, and entirely in private. An analyst uploads the due-diligence agenda and the interview transcript through a simple form and gets back a finished, structured findings report, without a transcript ever leaving the firm's systems:

• Reads whatever it's given — agenda decks and transcript documents are converted to clean text automatically, whatever format they arrive in, and the agenda's questions are pulled out in full, including their sub-questions.
• Finds every answer — and checks it — for each agenda question, the system locates the answer in the transcript, then runs a second, independent AI pass that verifies the answer is genuinely supported by what was said. Answers are graded for reliability rather than taken on faith, so the findings don't rest on something the AI invented.
• Catches what wasn't asked — separately, it scans the whole transcript for things that were answered spontaneously but never appeared on the agenda — exactly the unprompted disclosures that can make or break a deal and that manual review tends to lose.
• Delivers a finished report, not raw output — the agenda answers and the unprompted findings are merged into a single structured spreadsheet, filed into the firm's existing document system, and shared as a secure, login-protected link by email.
• Private by design — the whole pipeline runs on the firm's own infrastructure, and every AI call goes to the firm's own private cloud region in its home country, so confidential deal data never leaves it. Stored data is encrypted, backed up daily, and access-controlled throughout.

The outcome

What used to mean an analyst combing through every transcript by hand is now an automated step — more consistent from one interview to the next, and with a safety net for the disclosures that surface without being asked. Because everything runs inside the firm's own cloud, that speed comes with a privacy posture strict enough for confidential deal work. The system has run in production since 2025, processing interviews end to end with daily backups.