systems running in production.
four engagements. client names redacted pending publication approval — reference calls available on request under nda. we publish numbers, not screenshots.
in production with a research desk for an enterprise advisory team.
a 4-person research desk spent two days per brief stitching together sources, transcripts and competitor scans. throughput hard-capped by headcount.
planner / executor / critic agent loop with custom tools for source ingestion, transcript search and citation auditing. human reviewers see the final brief, not the steps.
in production with a revenue team at a series-b b2b saas.
sdrs spent 65% of their time on dial time and discovery calls that didn't qualify. inbound funnel had no after-hours coverage.
low-latency voice agent handles inbound + outbound qualification and books demos. escalates to a human sdr on real intent signals. crm-native; reps see the call summary, not raw transcripts.
in production for six months at a mid-market saas.
tier-1 ticket volume scaled past the support team's capacity. previous chat tool hallucinated policy answers and required a docs rewrite for every product change.
hybrid retrieval (bm25 + dense + rerank) over docs + ticket history with strict citation gates. zero answers without sources. eval harness blocks regressions in ci.
in production with a defi protocol's treasury committee.
treasury operations lived in spreadsheets and manual signing rounds. signers had no live view of exposure; reporting lagged events by weeks.
audited solidity contracts for payouts + custom indexer feeding a real-time dashboard. signers see exposure live; ops runs in under an hour per week.
more coming as we ship in 2026.
active engagements remain in stealth until clients sign off on publication. expect new case studies as they clear approval.