Fit Money Coach

Upload any methodology as a PDF and get a personalized 30-day curriculum — grounded strictly in your source, tracked with real-time progress. No invented content, no generic templates.

Comment

Inspiration

90% of self-paced learners drop out before finishing a course. Not because the content is bad — because raw information is overwhelming, generic, and impossible to act on 15 minutes at a time. Corporate wikis, coaching PDFs, and training manuals all suffer from the same problem: they're built to be read, not built to be followed.

We asked: what if any methodology — a course, a coaching framework, an internal SOP — could turn itself into a personalized, day-by-day curriculum the moment someone uploads it?

That question became Fit Money Coach.

What It Does

Fit Money Coach is an AI-native curriculum builder. A user uploads any methodology — a PDF or pasted text — describes the learner (experience level, goal, time available, biggest obstacle), and the platform instantly returns:

Output Description
4-Week Curriculum A structured 30-day plan broken into weekly themes
Daily Micro-Modules 15–20 minute sessions with a clear objective and summary per day
Source-Grounded Content Every module traces back to the uploaded methodology — nothing invented
Live Progress Tracking Checkboxes and a real-time progress bar as the learner completes each day

We used our own Fit Money Circle financial discipline framework as the proof-of-concept source document — proving the tool on a real methodology, not a toy example.

How We Built It

  • Stage 1 — Source Ingestion: The user pastes text or uploads a PDF, parsed client-side and passed as raw context — no external RAG layer needed for a single-document, single-session use case.
  • Stage 2 — Structured Plan Generation (openai/gpt-5.6-so1): Source content and learner profile go to openai/gpt-5.6-so1 via the Lovable AI Gateway with strict Structured Outputs — four weeks, 5–7 modules each, enforced through prompt-level determinism (no temperature parameter, since the gateway rejects it for this model).
  • Stage 3 — Interactive Rendering: The returned JSON maps directly into expandable weekly sections, per-day cards, functional checkboxes, and a real-time progress bar.

Tech Stack: Lovable (React + Tailwind) · Lovable Cloud (Supabase-backed) · OpenAI gpt-5.6-so1 via Lovable AI Gateway · Custom "Warm Premium" palette (cream, terracotta, sage)

Challenges We Ran Into

  1. Claude wasn't available through the gateway: We planned to run generation through Claude, but Lovable's AI Gateway only routes openai/* and google/gemini-* models. We pivoted to openai/gpt-5.6-so1 with Structured Outputs — the stronger choice for guaranteed schema conformance anyway.
  2. No temperature control: GPT-5 family models reject a custom temperature through the gateway (400 error). We solved this with prompt-enforced determinism instead.
  3. Occasional malformed JSON: The model sometimes wrapped output in markdown fences. Solved with strict "output ONLY valid JSON" instructions plus a one-time automatic retry on schema failure.
  4. Getting the palette right: Our first instinct was dark and corporate. But finance and habit change are emotionally loaded — we rebuilt around warm cream and sage to make the tool feel like a coach, not an algorithm.

Accomplishments We're Proud Of

  • ✅ Full pipeline working end-to-end — upload → Structured Output generation → interactive 30-day plan, live in production
  • ✅ Zero invented content — every module explicitly grounded in the source document
  • ✅ Real-time progress tracking — functional checkboxes, dynamically updating progress bar
  • ✅ Deterministic JSON at scale — strict Structured Outputs plus retry logic means the UI never breaks
  • ✅ Built and shipped in under 48 hours — blank canvas to tested, working app

What We Learned

Model availability isn't universal — checking the actual supported catalog before writing prompts would have saved a pivot. Once corrected: the "best" model on paper isn't always the right tool for the job. Structured Outputs fit this task better than a general-purpose reasoning model would have, because the deliverable needed guaranteed schema conformance more than creative flexibility.

We also learned visual tone isn't cosmetic for emotionally-loaded subjects like personal finance — the palette measurably changes how safe a tool feels before a word is read.

What's Next for Fit Money Coach

  • Immediate: A "Dark Premium" commercial fork for the Fit Money Circle brand, plus coach/client authentication.
  • Q3 2026: Dynamic single-input ingestion (in the style of our Market Scout AI project), white-label deployment for coaches and agencies.
  • Q4 2026: B2B onboarding module for agencies converting internal SOPs into automated training, usage-based pricing.

Fit Money Coach is the onboarding brick. The methodology becomes the curriculum. The curriculum becomes measurable behavior change. That's the Medusa Black Labs flywheel.

Built With

Share this project:

Updates