Archetype Code
An LLM pipeline generating multi-layered personal reports from several data sources: five independent computed layers are synthesized by the model into a single structured portrait. FastAPI + PostgreSQL + Next.js, free basic result and a paid full report.
Table of Contents
An experiment at the edge of symbolic systems and LLMs: can you assemble a coherent, verifiable portrait from an «esoteric» input and ship it as an engineering product rather than «magic».
Context
People get stuck in repeating patterns and look for language to describe themselves — sharper than «you're an introvert». Archetype Code takes a name and birth date and returns a mythological archetype: decision-making mechanics, strengths, and «that's not me» markers.
How it works
Under the hood is a pipeline of five layers. Each gives its own slice, and the final layer fuses them into one portrait:
- Numerology — birth-date codes.
- Astrology — sign, elements, planets.
- Psychomatrix — Aleksandrov's method, 9 characteristics.
- Tarot — tasks for the current period.
- Narrative synthesis — the LLM assembles the scattered signals into a coherent archetype rather than dumping a list of «you have a 5».
The engineering frame
- Each layer is a separate deterministic computation; the «magic» lives only in the narrative, and it's verifiable.
- The LLM writes the portrait from a structured input rather than inventing facts from scratch — fewer hallucinations, a repeatable tone.
- Calibration on 400 public figures with verified dates — to check, not to take on faith.
Plans
- Free: archetype with a name and image, decision-making mechanics, «I'm in my element» and «that's not me» markers.
- Report (390 ₽): false scenarios and development paths, financial and relationship patterns, period tasks via tarot, practices.
What I took away as an engineer
- Symbolic input + deterministic layers + LLM synthesis = a managed loop, not a black box.
- The hard part isn't the math — it's the tone: the portrait must sound about the person, not about a «number of fate».
- Verifiability (400 figures) turns «esoterica» into a product you can debug like any other service.
Similar projects
Projects with similar technologies and tasks
Tech Path Finder
Educational platform for IT professionals with quizzes, mock interviews, code review, and personalized recommendations based on a knowledge decay algorithm
- Python
- FastAPI
- PostgreSQL
- Redis
- Kafka
- +5
Aginx
Automated LLM pipeline: brief → script → media generation → cut, one loop without manual intervention. An LLM (Claude) writes the scene-by-scene script with a hook and voice direction; a Python/asyncio pipeline generates video, voice, music and subtitles via fal.ai, Replicate and ElevenLabs and stitches a finished vertical video up to 60 seconds with ffmpeg.
- Python
- asyncio
- Typer
- Pydantic
- Claude
- +4
Slot‑Me.ru — Meeting Booking Platform
Cal.com for the Russian market: from architecture to production. FastAPI + React, FSD, OAuth, calendars, email, 196 tests.
- FastAPI
- React
- PostgreSQL
- Redis
- TypeScript
- +3