Contacts and chat RAG

A forensic chat analysis tool built on a Retrieval-Augmented Generation (RAG) pipeline. It ingests structured phone book and chat session data, embeds them into a vector store, and answers natural-language questions grounded in the evidence.

Architecture

Two-layer design:

• Generic RAG (Services/, Interfaces/, Models/) -- reusable embedding, vector search, and streaming answer generation
• Domain layer (Domain/) -- forensic-specific logic: participant resolution, aggregate queries, temporal filtering, pre-computed forensic facts

Design Patterns

• Content-addressable caching -- SHA-256(filename + content) keys; skips OpenAI embedding calls when files haven't changed
• HyDE (Hypothetical Document Embeddings) -- generates a draft answer before embedding for better semantic retrieval
• Decorator pattern -- DomainRagOrchestrator wraps RagOrchestrator with domain enrichment
• Adapter pattern -- OpenAiEmbeddingAdapter bridges OpenAI SDK to the generic IEmbeddingGenerator interface
• Polly resilience -- exponential backoff with jitter for transient API failures (429, 5xx)
• Atomic file writes -- cache writes go to .tmp then File.Move to prevent corruption

Prerequisites

• .NET 10 SDK
• OpenAI API key with access to chat and embedding models

Setup

1. Clone the repository
2. Create a .env file in the project root:

   OPENAI_API_KEY=sk-...
   

3. Place forensic data files:
   • phonebook.txt -- raw phone book export
   • chats/ -- raw chat export directory

Usage

Interactive mode:

dotnet run

Commands in interactive mode:

Command	Description
exit / quit	Close the application
exit with save	Save conversation history as JSON and exit
/clear	Reset conversation context
/help	Show help
Ctrl+C	Cancel the current query

Running tests:

dotnet test                                          # all tests
dotnet test --filter "FullyQualifiedName~Unit"       # unit only
dotnet test --filter "FullyQualifiedName~Functional" # functional only
dotnet test --verbosity normal                       # verbose output

Validation test suite:

dotnet run -- --test

Runs 28 automated test cases covering:

• Phone book queries (contacts, phone numbers, chat file IDs)
• Temporal filtering (monthly activity, busiest day)
• Aggregate queries (all contacts in a period, cross-file counts)
• Comparison and exclusion queries
• Miss attribution detection
• Single-chat-max and timeline edge cases

Improvement TODO:

1. Linux
2. Page Index
3. GraphRAG
4. RAPTOR
5. MCP (drop interactive CLI)
6. C.R.U.D data for RAG data
7. Mutation tests
8. Line coverage minimum 95%, mutation coverage minimum 95%
9. Dockerfile