Company Intelligence (.NET)

RAG-based document intelligence assistant. Upload PDFs, ask questions offline with grounded, cited answers. .NET reimplementation of company_intel.

How it works

Phase 1: INGEST (upload)              Phase 2: QUERY (chat)
───────────────────────────           ────────────────────────
Upload PDF document                   "What does the report say about X?"
  → PdfPig extracts text                → Vector search (Qdrant)
  → Recursive chunking (128 tokens)     → Retrieve relevant chunks
  → Embed (all-minilm, 384-dim)         → LLM generates grounded answer
  → Upsert to Qdrant                    → Citations from stored knowledge

Example

Document ingestion	Chat — Q&A with citations

Tech stack

Layer	Choice
LLM	Llama 3.1 via Ollama (local-only)
Embeddings	all-minilm (384-dim)
Vector store	Qdrant
Agent framework	Microsoft.Agents.AI + AG-UI protocol
AI abstractions	Microsoft.Extensions.AI
Frontend	CopilotKit + Next.js 15
Orchestration	.NET Aspire
PDF extraction	PdfPig
Observability	OpenTelemetry → Aspire dashboard

Quick start

# Prerequisites: .NET 10 SDK, Node.js 20+

# Start everything
dotnet aspire run

This starts Ollama (+ model pulls), Qdrant, SQLite, API, and Next.js UI via .NET Aspire. Open the Aspire dashboard link from terminal output.

Project structure

src/CompanyIntel.AppHost/          → .NET Aspire orchestrator
src/CompanyIntel.Api/              → ASP.NET Core backend (RAG agent, ingestion)
src/CompanyIntel.UI/               → Next.js 15 frontend (CopilotKit + AG-UI)
src/CompanyIntel.ServiceDefaults/  → Shared OpenTelemetry, health checks, resilience
tests/                             → Integration & RAG evaluation tests
slides/                            → Marp presentation (GitHub Pages)

Presentation

View slides