PROJECT BRIEF / 07
Zero to one, end to end.
Axolytica, an AI-native social platform prototype
The context
Most enterprise AI work starts with a customer and a mandate. This venture started with a blank page: a thesis that a social network could be synthetic by design, with every post transparently AI-made and a profile shaped by a person's health and lifestyle signals. It was built independently from September 2025 to July 2026, with more than 280 commits.
The approach
- Define the product thesis and a roadmap that separates what is available today (health insights and personalization reports) from what is still coming (avatars, a social graph, messaging).
- Build the full stack independently: a FastAPI service, Supabase for authentication and data, and a static multi-page web app on Vercel.
- Use each technique where it fits: a scikit-learn pipeline for structured health inputs, LLMs for narrative reports parsed into structured JSON, and plain code for calculations.
- Prototype matchmaking with deterministic feature vectors and cosine similarity before paying for semantic embeddings.
- Capture early demand with a waitlist, an About narrative, and a collaboration form, and keep one codebase from laptop to production.
System architecture & flow
- 01
Intake & identity
Health and lifestyle intake with Supabase authentication and per-user records.
- 02
Models & reports
A scikit-learn risk pipeline and LLM-written coaching and personalization reports, parsed into structured JSON.
- 03
Matching baseline
Deterministic feature vectors and cosine similarity recommend partners, a cheap baseline before semantic embeddings.
- 04
Trust & release
JWT-verified endpoints, restricted CORS, reCAPTCHA, dev-only shortcuts gated by environment, and the same code from local to production.
The entrepreneurial connection
Entrepreneurial, founder-style ownership: shaping an idea into a scoped product, making build and trust decisions early, and putting something real in front of the public. It complements the architect's view with the experience of carrying a product from a blank page to a public early-access page.
What to look at
Start with the thesis
Read the early-access page to see how the product is positioned and what is deliberately marked as coming soon.
Ask for the walkthrough
The repository is private. A walkthrough covers the API, the data model, the ML pipeline, the LLM report flow, and the matching baseline.
Judge the scope honestly
Treat it as a prototype. The health model is illustrative, matching is a baseline, and nothing here is clinical guidance.
Continue exploring