siddhi0808/groundtruth_ai

How siddhi0808/groundtruth_ai works

GroundTruth AI is a document-grounded hallucination detector. A user submits a question and an LLM answer through the browser; the API retrieves evidence, asks a judge for a verdict, returns the result, and records detection history. Documents enter through upload or batch ingestion, then are extracted, chunked, embedded, and indexed. The product also exposes health, statistics, history, and dashboard views. The README documents both vector search and degraded SQLite keyword search; unsampled API details are represented only where documented.

Main entry points: main.py API handlers · ingest.py Batch ingestion · script.js UI client · database.py Chunk and history store

Components 16

User experience

Detector UI frontend/index.html
Dashboard view frontend/dashboard.html
History view frontend/history.html
UI client frontend/script.js

Detection API

API handlers backend/main.py
Runtime config backend/config.py

Knowledge indexing and retrieval

Batch ingestion backend/rag/ingest.py
Document extraction backend/rag/document_loader.py
Token chunking backend/rag/chunker.py
Embedding model backend/rag/embeddings.py
Evidence retrieval backend/rag/retriever.py

Verdict and persistence

Verdict judge backend/llm/judge.py
Chunk and history store backend/database.py

Other components

User
Browser
Ollama

Connections 27

User → Browser: uses README.md
Browser → Detector UI: opens README.md
Detector UI → UI client: runs frontend/script.js
UI client → API handlers: submits detection frontend/script.js
API handlers → UI client: returns verdict README.md
API handlers → Evidence retrieval: retrieves evidence backend/main.py
API handlers → Verdict judge: judges answer backend/main.py
API handlers → Chunk and history store: records detections backend/main.py
API handlers → Document extraction: extracts uploads README.md
API handlers → Batch ingestion: indexes uploads backend/main.py
Batch ingestion → Document extraction: loads documents backend/rag/ingest.py
Batch ingestion → Token chunking: chunks text backend/rag/ingest.py
Batch ingestion → Embedding model: embeds chunks backend/rag/ingest.py
Batch ingestion → Chunk and history store: inserts chunks backend/rag/ingest.py
Token chunking → Embedding model: measures tokens backend/rag/chunker.py
Evidence retrieval → Embedding model: embeds query backend/rag/retriever.py
Evidence retrieval → Chunk and history store: searches chunks backend/rag/retriever.py
Chunk and history store → Runtime config: reads settings backend/database.py
API handlers → Runtime config: reads settings backend/main.py
Verdict judge → Runtime config: reads settings backend/llm/judge.py
Verdict judge → Ollama: requests judgment backend/llm/judge.py
API handlers → Dashboard view: serves page backend/main.py
API handlers → History view: serves page backend/main.py
Browser → Dashboard view: opens README.md
Browser → History view: opens README.md
UI client → API handlers: requests upload frontend/script.js
UI client → API handlers: requests history and stats README.md

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·