My Black :[ROOTS] Dataset Builder

My Black :[ROOTS] Dataset Builder is a browser-based system designed to help users preserve, explore, and analyze George Freeney Jr's family heritage while building high-quality datasets for AI model fine-tuning. It combines structured genealogy, recursive relationship tracing, memory-aware AI responses, and automated dataset generation for open-source models such as gpt-oss:20b and gpt-oss:120b.

NOTE:

ALWAYS refer to, and review the repo's current supabase.sql file for this project.

The platform integrates a full Supabase PostgreSQL schema, recursive SQL views, pgvector embeddings, cultural feedback loops, and two AI agents:

• Agent 1 answers genealogy questions using database tables and views + chat history, feedback memory, and embedding tables.
• Agent 2 builds curated training examples using genealogy data, feedback, and past interactions.

This repository contains the full SQL bootstrap file, schema, and application logic needed to deploy the Dataset Builder.

---

🚀 Features

Genealogy Management

Uses structured tables and recursive views to explore:

• Ancestry relationships
• Descendants
• Siblings, aunts/uncles, nieces/nephews, cousins
• Haplogroup tracing
• Enslaved ancestry tracing
• DNA match indicators
• Stories, alternate names, suffixes, approximate dates

AI Dataset Generation

Automatically builds dataset examples for:

• AI fine-tuning
• Retrieval-augmented tasks
• Cultural reasoning and accuracy
• Question/answer examples from traces and relationships

Dataset entries are stored in:

• Supabase (dataset table)
• Optional Google Sheets mirror
• Vector embeddings table (embeddings) using pgvector

Self-Improving Feedback Loop

The feedback and chat_history tables allow the system to:

• Record user ratings
• Improve future genealogy answers
• Improve dataset generation
• Track cultural corrections
• Provide long-term memory for both agents

---

📦 Full SQL Schema

The database includes all tables, views, extensions, triggers, functions, and RLS policies required for the Dataset Builder.

Core Tables

• persons
• marriages

AI Tables

• dataset
• feedback
• chat_history
• embeddings (pgvector, 768-dimensional vectors)

Relationship Views

• ancestry_relations
• descendants_relations
• lateral_relations

Trace Views

• paternal_haplogroup_trace
• maternal_haplogroup_trace
• paternal_enslaved_trace
• maternal_enslaved_trace

Functions

• update_timestamp()
• enforce_single_home_person()

Triggers

• Auto-update timestamps for:
  • persons
  • marriages
  • dataset
• Enforce single home person
• Dataset timestamp behavior

Extensions

• uuid-ossp
• vector (pgvector)

Row Level Security (RLS)

• RLS enabled on:
  • persons
  • marriages
  • dataset
  • feedback
  • chat_history
  • embeddings
• Public read-only for genealogy
• Authenticated insert/update for AI tables

A complete SQL bootstrap file is included in the repository and can recreate the entire schema without data loss.

---

🧠 AI Agents

Agent 1 — Genealogy Q&A

• Uses all views + persons & marriages
• Leverages the feedback table as memory
• Answers questions such as:
  • “List my paternal enslaved ancestors.”
  • “Show my maternal cousins.”
  • “Trace my haplogroups.”

Agent 2 — Dataset Builder

Uses:

• dataset
• feedback
• chat_history
• persons
• marriages
• all relationship views

Produces:

• curated query/response examples
• anonymized person-based training data
• context-aware pairing examples
• improved versions of incorrect responses

Supports Ollama and OpenAI gpt-oss models.

---

📊 Dataset Workflow (n8n Optional)

The system can integrate with n8n to:

1. Receive a webhook (feedback_id or manual refine request)
2. Pull feedback, chat_history, and related genealogy context
3. Generate 3–10 dataset rows
4. Append rows to:
   • Supabase dataset
   • Google Sheets
   • embeddings (vector representation)
5. Return status updates to the app

This allows continuous dataset improvement.

---

🛠 Development

Requirements

• Supabase project
• pgvector extension enabled
• SQL bootstrap applied
• Local or deployed frontend
• Optional n8n cloud account for dataset automation

SQL Bootstrap

Located in:
/sql/full_bootstrap.sql

Running it will create:

• All tables
• All views
• All functions
• All triggers
• All RLS
• All indexes
• Embeddings table
  with no destructive actions.

---

🔒 Privacy

All genealogy, feedback, datasets, and embeddings remain in the user’s private Supabase instance. Nothing leaves the user's database unless they explicitly export it.

---

📁 Repository Structure

/
│── sql/
│     └── full_bootstrap.sql
│
│── src/
│     ├── components/
│     ├── agents/
│     ├── views/
│     └── utils/
│
│── README.md
└── package.json

---

📬 Contact

This project is part of the My Black :[ROOTS] ecosystem and supports ongoing development of culturally accurate AI tools for Black American genealogy and historical preservation.