Enterprise-AI-Research-Agent

┌─────────────────────────────────────────────────────────────────────┐
│ USER INTERFACE (Explainability Surface)                             │
│ Streamlit (MVP) / Next.js later                                     │
│ • Research topic input                                              │
│ • Live pipeline progress (which stage is running)                   │
│ • Expandable sources, findings, classifications, contradictions     │
│ • Final report with clickable citations + confidence                │
│ • Session history + knowledge-base search                           │
└──────────────────────────────┬──────────────────────────────────────┘
                               │ HTTP / SSE (streaming)
┌──────────────────────────────▼──────────────────────────────────────┐
│ APPLICATION / API LAYER (Maintainability Boundary)                  │
│ FastAPI                                                             │
│ • POST /research → launches pipeline, streams every node            │
│ • GET /sessions, /session/{id}, /knowledge/search                   │
│ • Clean contracts only — UI never talks to agents or DB directly    │
└──────────────────────────────┬──────────────────────────────────────┘
                               │
┌──────────────────────────────▼──────────────────────────────────────┐
│ AI INTELLIGENCE LAYER (Scalable + Explainable Core)                 │
│ LangGraph StateGraph (Anthropic-style orchestrator-worker)          │
│                                                                     │
│ 1. Planner (Define Research Questions)                              │
│ 2. Searcher / Collector (parallelisable)                            │
│ 3. Store Sources                                                    │
│ 4. Extractor                                                        │
│ 5. Comparator + Classifier                                          │
│ 6. Contradiction Detector                                           │
│ 7. Synthesizer (Generate Conclusions)                               │
│ 8. Traceability Writer                                              │
│                                                                     │
│ Shared typed State + Checkpoint after every major node              │
│ LLM = Groq free-tier (Llama-3) or Ollama local (tool-calling)       │
└──────┬────────────────────────────────────────────┬─────────────────┘
       │                                            │
┌──────▼────────────────────┐             ┌─────────▼─────────────────┐
│ DATA & KNOWLEDGE LAYER    │             │ EXTERNAL RESEARCH / DATA  │
│ (Persistent & Reusable)   │             │ DuckDuckGo / Tavily free  │
│ • SQLite (structured      │             │ or local SearXNG          │
│   lineage & audit trail)  │             └───────────────────────────┘
│ • ChromaDB (vector        │
│   embeddings of sources   │
│   + findings)             │
│ Survives restart.         │
│ Grows with every run.     │
└───────────────────────────┘