From 90 Days to Minutes
Skills Platform
Recorded and edited by me. It demonstrates an AI system; it wasn’t made by one.
Customers spent 90 days trying to ship a conversational AI. I championed and drove creation of a modular conversation orchestration platform (engineering implemented core platform), reducing time-to-POC from ~90 days to minutes for template deployments. When LLMs arrived, the architecture was ready.
Context
If you know LangChain or LangGraph, this is the same shape of problem, solved earlier and under real-time constraints: route a conversation's input and output through whichever provider should handle it — generative, NLU, or plain deterministic — so GPT-3 could run alongside Dialogflow, Watson and Lex, with translation and input/output checks addable at any step without writing code. Comparable in execution, but for live conversation, where a stalled turn is the failure.
Problem Space
90+ day time-to-first-conversation, every deployment bespoke, no reusable components, Soul Machines excluded conversation design from scope.
What I Did
Championed from three roles, built prototypes instead of decks, navigated C-suite alignment.
What I Built
- Skills Platform orchestration
- No-code assembly
- Elegant failure skill
- Generative Conversation Skill (GPT/RAG)
Adoption
99% adoption, 6,300+ skills, 1,400+ customer-created.
Outcome
- ~90 days to minutes for template POCs
- Foundation for all LLM integration
- Patent filed January 2021 — PCT WO2022149076A1, US US20240054292A1, national phase across six jurisdictions
Why This Matters
Built the thing that let everyone else build things. Architecture from 2019-2020 enabled seamless LLM integration in 2023+.