A high-performance, minimalist AI workspace designed for deep research and multi-agent orchestration.
In the age of LLMs, the problem isn't getting answers—it's information overload and context fragmentation. Researchers spend more time managing tabs and copy-pasting than actually thinking. Lexicon AI solves this by creating a "Single Source of Truth" where AI agents work for you inside isolated, context-aware workspaces.
Lexicon uses a Multi-Agent Orchestration pattern:
- The Researcher: Scans uploaded documents and live web data.
- The Analyst: Synthesizes findings and identifies contradictions.
- The Writer: Drafts reports into the integrated "Insights" markdown editor.
- True Workspace Isolation: 100% separate vector databases and chat histories per project.
- Agentic Status Stream: Real-time visualization of agent "thought processes" using Server-Sent Events (SSE).
- Offline-First Voice AI: Native "Listening" mode for hands-free research interaction.
- Zero-Boilerplate RAG: High-precision citations mapped directly back to source PDFs.
- Frontend: Next.js 14, TypeScript, TailwindCSS (Professional Neutral Theme)
- Backend: FastAPI (Asynchronous Python 3.10+)
- Agentic Framework: CrewAI / LangChain
- Intelligence: Groq (Llama 3.3), Ollama (Local Support for privacy-first research)
- Storage: Vector Store (ChromaDB), JSON-based persistence for lightweight workspaces.
- Start the Frontend:
cd frontend && npm install && npm run dev - Start the Backend:
cd backend && pip install -r requirements.txt && uvicorn app.main:app --reload
MIT License. Created by Bhoompally Kalyan Reddy for the NSoC'26 Competition.