All case studies
Matchmaking · Discovery
A matchmaker got a repeatable pipeline for sourcing candidates from public rosters
A head-to-head benchmark across 5 AI systems on 69 real candidates produced a copy-paste pipeline plus a hallucination-detection guide.
The problem
A matchmaker needed a repeatable workflow: given a public candidate page (a university athletic roster, a professional directory), find each woman's LinkedIn, Facebook, and Instagram profiles and produce a structured spreadsheet with photos. Question: which AI tool is best at this?
What we built
- Cross-tool benchmark across Claude Sonnet 4.6, Claude Opus 4.6, Perplexity Pro Deep Research, Manus AI, and Gemini
- Tested on 69 real candidates from two public rosters
- Self-contained benchmark report (HTML) with side-by-side accuracy scores
- Recommended Best Combined Pipeline: Sonnet for scrape → Perplexity for LinkedIn → Manus for Instagram and Facebook → Perplexity 2nd pass for zero-link → Sonnet to merge
- Projected coverage: LinkedIn 75-80%, Facebook 25-30%, Instagram 65-70%, at least one link 90%+
- Hallucination detection guide flagging specific failure modes (Gemini fabricates LinkedIn slugs with fake hex IDs)
Stack
Outcome
Benchmark complete. Recommendations delivered and in production use.