Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Building a Chat-First AI Sales Cockpit Solo — The Anti-Hallucination Lesson
Learn how to build a chat-first AI sales cockpit that structurally prevents hallucinations by grounding LLM calls in retrieved context, surfacing gaps as verifiable tags.
A working chat-first AI execution cockpit for enterprise sales. Sellers operate in one conversational surface all day — outbound, replies, meeting prep, coaching — instead of switching between six tools. Built solo by an SDR who lived the friction.
The core insight is structural anti-hallucination. Every LLM call is grounded in retrieved context. The model is never asked to recall specifics from memory. Gaps surface as [VERIFY:] tags rather than fabricated facts. This isn’t a prompt instruction — it’s an architecture pattern that makes the output trustworthy enough for customer-facing drafts.
Live demo on a fully simulated dataset. I’ll walk through the architecture (4-band priority queue, self-feeding sequence engine, overnight email triage, reply drafter with grounded retrieval), and the one reusable lesson: why prompt-based “don’t make things up” fails, and how to fix it structurally