FDD Update Engine: AI + Deterministic Drafting for Franchise Disclosure Documents

I was hunting for an untapped opportunity where LLM document synthesis and analysis could do real work that nobody had automated yet. Franchise Disclosure Documents (FDDs) was the biggest wedge I found. Every U.S. franchisor — roughly 9,000 brands — has to update its 23-item FDD every year, it’s done by hand by attorneys at $4,000–$15,000 a filing, and no one is automating the franchisor side. I also have a government angle to lean on (Secret Clearance through ANG), which matters for the larger adjacent wedge (defense / CUI compliance documents). The full market reasoning lives in the venture assessment.

And So I built a demo to figure it out, and while building it changed the idea.

The original plan was 23 extraction prompts: read each item out of the prior FDD and re-generate the text from scratch. What I realized is that’s not how the work actually happens. An attorney updating an FDD doesn’t draft 23 items from a blank page — they put last year’s FDD next to this year’s source documents and compare them section by section, fixing what changed and keeping it compliant.

So I pivoted. The real product is simpler and matches the actual workflow: upload last year’s FDD, upload the new inputs (amended franchise agreement, outlet roster, updated data), and get a side-by-side, tracked-changes comparison the attorney can review and finalize — instead of writing from nothing. The deterministic items (9, 17, 20, 23) are generated in pure Python; the narrative items are diffed by an LLM with the prior text routed per item.

Retired

This was a completed experiment, not a current AkanmuTech product. The hosted demo and its paid runtime were retired in September 2026. This write-up remains as an archive of what was built and learned.

• The venture assessment behind it → — the market, the incumbents, the unauthorized-practice-of-law risk, and why I led with FDD.
• Code on GitHub →