🔬 Deep Fusion Research

AI‑powered workflow for turning existing literature into new, cross‑domain insights.

deepfusion.mp4

---

✨ What it does

1. Ingest any paper / URL – PDFs are parsed with Mistral; web articles with Firecrawl.
2. Generate a structured research plan with LangGraph state machines & LLMs.
3. Realtime SSE stream shows every reasoning step as the graph executes.
4. User‑selected Focus items (notes, key findings, new hypotheses) are merged back into the agent context.
5. Fusion phase re‑runs the graph to produce a "next‑step" report – undiscovered angles, converging domains, or follow‑up experiments.

---

🚀 Quick start

pnpm i

cp .env.example .env.local && $EDITOR .env.local

pnpm dev  # ➜ http://localhost:3000

Minimal prod build

pnpm build && pnpm start

Required env vars

Key	Used for
OPENAI_API_KEY	Chat completions (planning & writing)
LANGSMITH_API_KEY	LangSmith traces (optional)
LANGSMITH_PROJECT	Project name in LangSmith
PERPLEXITY_API_KEY or TAVILY_API_KEY	Web search
FIRECRAWL_API_KEY	Article scraping
MISTRAL_API_KEY	PDF extraction

---

🏗️ Tech stack

• Next.js 14 / React 18 – App router, Server Actions, Tailwind CSS
• LangGraph + LangChain – Deterministic state machine orchestration
• SSE for live step streaming
• Zustand & shadcn/ui for UX

---

🛣 Roadmap

• Vector store & RAG for very large corpora
• Collaborative cursors & comments
• 1‑click export → Notion / Markdown / PDF

---

📄 License

MIT © 2025 Prasanna A P