MarketFramer
The pain
Analysts at banks and funds spend days pulling numbers out of 10-Ks and investor decks and wiring them into valuation models by hand — slow, error-prone work that has to be redone every quarter.
What we built
An AI engine purpose-built for financial modeling. Point it at a company and it generates institutional-grade deliverables — DCFs, LBOs, comps, company profiles — as real, auditable Excel and PowerPoint files, sourced straight from filings.
Where it landed
We landed one early partner, a firm we generated valuation models for directly, and cut their turnaround from days to under an hour. It never grew beyond that first partnership.
DocSlicer
The pain
A problem we hit ourselves while working with 10-Ks, 10-Qs, and other SEC documents for MarketFramer. Off-the-shelf splitters cut documents at arbitrary token boundaries or formatting cues — slicing sections mid-sentence, flattening hierarchy, and mangling tables — so the model retrieves fragments with no context.
What we built
A structure-aware chunking service that detects a document’s hierarchy and keeps every chunk scoped to its section path, with tables and code blocks intact. Delivered as a Python SDK, REST API, and web app.
Where it landed
Benchmarked against fixed-token, recursive, semantic, and Docling splitters across legal, financial, and technical corpora, structure-aware retrieval reliably surfaced the right passage — but the space was crowded and most teams were content to ship with generic splitters, so pull stayed thin.
Qualigon
The pain
A “deviation” is any time something in a manufacturing process doesn’t go as planned — a machine reading is off, a step runs long, a temperature drifts out of range. Deviations can block production and cost millions, so investigating and documenting them properly, per FDA regulations, is critical — and often painfully slow.
What we built
AI that guided investigators through root cause analysis, pulled data from connected systems, and drafted audit-ready reports — cutting investigation time from days to about an hour.
Where it landed
Demoed to many QA specialists and deviation managers, with consistently positive feedback as we iterated — but the blocker wasn’t technical. It was getting in front of the right decision-makers and convincing them to commit to a partnership.