Mirrors — Goal Tracking + AI

Overview

Fullstack + AI engineering for a goal-tracking startup MVP. Web app pairing user-defined goals with LLM-driven check-ins and coaching.

This project is currently on pause while the founding team re-assesses direction. I can walk through what shipped and what was learned.

Challenge

The founding team wanted to validate whether goal-setting paired with LLM coaching could sustain a habit loop. They needed an end-to-end MVP quickly — real product, real users — without stapling ChatGPT into a landing page and calling it done. No vibe-coding, no fake demo.

What I shipped

• Built the full web-app skeleton — auth, data model, user flow — before any AI wiring, so the LLM was a feature, not the product.
• Modelled goals and check-ins in Postgres with a schema that supported both freeform reflection and structured milestones.
• Integrated LLM tool-calling for coaching prompts, with typed function signatures instead of raw completion parsing.
• Prototyped RAG over the user's own history so check-ins could reference past state without leaking context to other users.
• Shipped an honest, testable MVP — currently on pause while the founding team re-assesses market direction.

Stack