We don't publish client names — every engagement runs under NDA. What we can publish: the shape of the problem, what we built, and the numbers it runs at today.
A second-hand goods marketplace needed continuous visibility into competitor pricing, user sentiment on short-video platforms, and local regulation changes — previously assembled by hand from screenshots and spreadsheets.
We built a scheduled collector fleet (browser-based and API-based), a deduplication layer keyed on content fingerprints, a classification pipeline, and an internal dashboard with alerting. Duplicate ingestion — the classic failure mode of "pull everything" crawlers — is blocked at the schema level with unique keys and fingerprint checks.
An investment team screening dozens of franchise brands per quarter was doing due diligence in spreadsheets: media sentiment, litigation records, store counts, trade-area data — every analyst collected their own version of the truth, and scores had no evidence trail.
We built a data platform that pulls news, short-video mentions, litigation feeds and store-location data into one model, then scores each brand on a rules-based rubric. The house rule we enforced: every score must show its evidence — chart, source list and data provenance for each dimension. Missing data is marked as such, never scored as "good".
Years of documents — contracts, proposals, meeting notes, images of whiteboards — lived in shared drives nobody searched. The company wanted one place where data is ingested, classified, and askable in natural language.
We delivered an ingestion pipeline (PDF/Word/Excel parsing, image OCR, AI auto-classification into project domains), a structured entry store with tags and permissions, and a retrieval layer. Export operations are fully logged: who took what data out, when, and why.
Before selling GEO audits we built the instrument: a harness that fires buyer-intent questions at ChatGPT, Perplexity, Gemini and a China-market engine, records whether the brand appears, who owns the slot, and which domains get cited — then tracks movement weekly.
This is the same pipeline that runs the audit product. It answers the only question clients actually have: "why does the AI never mention us — and what changes that, in what order?"
Tell us what's broken. One paragraph is enough — we'll reply with a scope and a price.
References and deeper case detail available under NDA on request.