project · 2025-2026
AmplifyAI, GenAI platform for CPG innovation
Co-founded an end-to-end GenAI platform that automates the innovation workflow for Consumer Packaged Goods companies, from idea prototype through to production-ready concept. Full-stack React + Python + LLM integration, shipped to early customers.
Co-founded with the goal of compressing the CPG innovation cycle, the multi-month process of going from “what if we made this product” to a production-ready concept, into something a small team can do in days.
What it does
The platform automates the workflow product teams already follow but at LLM speed:
- Idea capture: structured prompt for the product hypothesis, target consumer, and constraints.
- Concept generation: LLM expands the hypothesis into multiple variants with claims, ingredients, packaging cues, and target shelf placement.
- Validation: each variant is scored against historical CPG datasets and trend signals.
- Iteration: rejected variants feed a refinement loop. Surviving variants get full ideation packages: copy, packaging mockup briefs, regulatory considerations.
- Hand-off: outputs land in formats that downstream design / ops / R&D teams can act on.
How it is built
- Frontend: React + TypeScript with a streaming UI so users see partial reasoning as the agent works.
- Backend: Python service orchestrating multiple LLM calls per workflow step, with deterministic post-processing where outputs need to be structured.
- LLM integration: provider-agnostic at the boundary so we can swap between Claude, GPT, and open-source models depending on the task.
- State: every workflow step persists; users can resume, branch, or compare across runs.
My role
Architected the system end-to-end, lead engineering, ship the product alongside the founding team. The interesting tradeoffs are not technical: they are about which steps deserve LLM autonomy and which deserve a human gate. We are still calibrating that.
Why this earns a spot in projects
Two reasons. One: it is the project where I am most directly responsible for product outcomes, not just engineering quality, and that perspective changes how I scope every problem. Two: it is a real test of “agentic systems in domains where the user is not technical.” The lessons here translate directly into how I think about agentic UI in my main role at TomTom.