The Deli Line Problem: Building an AI Pipeline for React Apps That Doesn't Fall Over

When your AI image analysis, search embeddings, and chat assistant all share one queue, you have a deli line problem.

I'll walk through how we built Ideate's multi-provider AI pipeline with Convex — priority queuing, circuit breakers, and adaptive batching — and how the same real-time architecture works across web (TanStack Start) and mobile (React Native) with shared business logic.

From Pixels to Meaning: Building an AI Content Pipeline That Actually Works

Every AI demo looks magical. Then you ship it and discover your image analysis returns "this is a photo" for 30% of uploads, your embedding model hallucinates categories that don't exist, and your users' searches return confidently wrong results.

I'm the CTO of Ideate, an AI-powered creative workspace for design teams. We built a content intelligence pipeline that analyzes images, generates semantic embeddings, and powers natural language search across visual libraries — using Gemini for vision and embeddings, Claude for conversational AI, and a custom orchestration layer on Convex.

This talk is about what broke, what we fixed, and what we learned shipping AI features to real creative teams:

- Why we validate every AI response against a strict schema before it touches our database — and the production incident that taught us to
- How we built adaptive batching that processes 5-15 images at a time (not the 50 I accidentally told my team) based on real-world
throughput limits
- The priority queue system that keeps interactive features fast while background analysis runs behind — our "deli line"
architecture
- Why retrieval-augmented search with reciprocal rank fusion outperformed pure vector search by 3x for creative content

No slides full of architecture diagrams you'll forget. I'll show real code, real failure modes, and the actual before/after of search results. You'll walk away with patterns you can apply to your own AI features Monday morning.

The Wrong Number: What Happens When Your CTO Guesses Confidently

I told my team our AI pipeline processed images in "batches of 50." The real number was 5. They built a sprint plan around my wrong answer, and nobody questioned it because I said it with confidence.

This talk is about what happens when technical founders optimize for sounding certain instead of being accurate. I'll share the specific moments where my unverified claims shaped real product decisions, the codebase audit I ran that same day to fact-check myself, and the pre-meeting system I now use before every architecture discussion.

You'll walk away with:
- A framework for calibrating technical confidence at the earliest startup stage
- A pre-meeting checklist you can use Monday morning
- Permission to say "let me verify that" and proof that it builds more trust than having all the answers

This talk is for first-time CTOs, tech leads, and anyone who's ever guessed a number in a meeting. I told my team our AI pipeline processed images in "batches of 50." The real number was 5. They built a sprint plan around my wrong answer, and nobody questioned it — because I said it with confidence.

This talk is about what happens when technical founders optimize for sounding certain instead of being accurate. I'll share the specific moments where my unverified claims shaped real product decisions, the codebase audit I ran that same day to fact-check myself, and the pre-meeting system I now use before every architecture discussion.

You'll walk away with:
- A framework for calibrating technical confidence at the earliest startup stage
- A pre-meeting checklist you can use Monday morning
- Permission to say "let me verify that" —and proof that it builds more trust than having all the answers

This talk is for first-time CTOs, tech leads, and anyone who's ever guessed a number in a meeting.

Adaptable to 5, 15, 20, 30, or 45-minute slots. Previously developed from real experiences at Ideate, an AI-powered creative workspace. First public delivery pending. Target audience: engineering leaders, startup founders, tech leads.

Code as a Canvas: What DJing Taught Me About Building Software

Before I was a CTO, I was a DJ. Before I was a DJ, I was burning CDs for my friends. Before that, I was a kid drawing on everything.

This talk is about what happens when you stop treating your creative life and your engineering career as separate things. I'll walk through five specific skills I learned from DJing and making art that directly shaped how I build software and lead a technical team from reading a room (user research) to mixing live (production debugging) to knowing when the crowd is done with a track (feature deprecation).

You'll hear real stories from the DJ booth and the deploy terminal, and walk away with a framework for finding the engineering lessons hiding in whatever weird creative thing you do on the side.

This talk is for anyone who's ever felt like their non-technical passions don't belong on their resume.

Adaptable to 5, 15, 20, 30, 45, or 90-minute (workshop) formats. Includes optional live DJ hardware demo with Pioneer DDJ-REV1.

Previously spoke at PhillyJS on real-time audio visualization. Target audience: developers, creative technologists, anyone with non-traditional paths into tech.