37 Shipped Systems
Body of Work
37
Production systems shipped
4
Celebrity digital twins
10.5 turns
Avg conversation depth (HelloSam)
Video Demo
37 production conversational AI systems shipped across 9 years. Brand ambassadors, celebrity digital twins, AI tutors -- built, authored, or materially improved, from conversation design through to production deployment.
Problem Space
Making people want to talk to AI a second time -- same core problem, different surfaces.
What I Did
Built, authored, or materially improved every agent -- conversation design through production deployment. Iterated across 9 years.
What I Built
- HelloSam (50%→75% success, 10.5 avg turns, 30K+ conversations)
- Digital twins (Mark Tuan, Francis Ngannou, Marilyn Monroe, Jack Nicklaus - 253K sessions)
- AI tutors (25K+ conversations/month)
- Name recognizer (10K names from 25 examples)
- STT benchmarking (discovered ~30% failure rate, designed conversations around it)
- 37 Digital People shipped
Adoption
37 production systems across brand ambassadors, celebrity digital twins, AI tutors, and enterprise deployments.
Outcome
- Time-to-first-response cut from ~4.5s to ~1.2s
- HelloSam became internal gold standard
- Each project informed the next
Why This Matters
Not separate projects -- iterations. Each one taught something that made the next one better.
Conversational AIDigital TwinsAI TutorsLatency OptimizationScale