Evidence-Bound Documentation

Enterprise Document Q&A with Verified Citations

Evidence-Bound is a RAG system for law firms where every answer must cite source documents — or the system refuses to answer. No hallucinations by design.

Getting Started

Getting Started — Clone to running in 10 minutes

Architecture

Architecture Diagrams — System overview, RAG pipeline, data model, deployment
Architecture Overview — Stack, security, providers, deployment tiers
Architecture Review — Honest gaps, competitive analysis, roadmap

Technical Reference

Technical Deep Dive — RAG pipeline internals, caching, cost tracking, PII
LLM Providers — Configure Azure OpenAI, Anthropic, Gemini, Ollama
Data Model — Full database schema (11 tables)
Provider Interfaces — LLM, Search, Embedding, Parser abstractions

Operations

Operations Runbook — Deploy, monitor, diagnose, rollback
Development Workflow — TDD, skills, hooks, review gates

Strategic

RAG Harness Spec — Extracting a reusable RAG framework
Multi-Tenant Readiness — SaaS gap analysis

Core Guarantee


If the system returns an answer:
  1. It cites a specific document, page, and character range
  2. The cited text exists verbatim in the source
  3. Confidence score is above threshold (default 0.70)
  4. Evidence grade (A/B/C) based on verification status

If it can't verify → it refuses. Every time.