Hermes AI Research Command Center

Cheetah Youth Program Project

Research Workflow · RAG · GPU Operations · AI Agents

A full-stack AI product that grew out of the Cheetah Youth Program and evolved into a local-first research command center connecting papers, semantic search, AI agents, reproduction tracking, and compute operations.

Program Context

This project was developed through the Cheetah Youth Program as a real-world product and engineering exercise rather than a standalone coding demo.

The program context pushed the project beyond interface design and into a complete delivery loop:

• translating activity requirements into product structure;
• designing frontend and backend boundaries;
• connecting structured data with real user workflows;
• integrating AI capabilities into a usable system;
• validating the system through iterative engineering and packaging.

The repository later evolved into Hermes AI Research Command Center, preserving the original product-engineering discipline while moving toward research tooling and AI-agent orchestration.

Why This Project Exists

Research work is usually fragmented across a PDF manager, browser tabs, SSH terminals, notes, model chats, and experiment logs. Hermes explores a different workflow: one command center that keeps papers, AI agents, compute resources, and reproduction records connected.

Core Workflow

Paper / Research Question
          |
          v
     Research Library
          |
     +----+-----+
     |          |
     v          v
 RAG Search   AI Agents
     |          |
     +----+-----+
          |
          v
Reproduction / Deployment
          |
          v
 GPU / Device Workspace

Main Capabilities

• Paper library — manage papers, PDFs, metadata, and structured analysis.
• Semantic research search — RAG-oriented search and paper conversations.
• AI command center — stream agent runs and tool events through a unified interface.
• Research agents — separate roles for research, paper analysis, and deployment workflows.
• Device workspace — manage remote compute resources and reproduction jobs.
• Reproduction records — keep implementation and experiment progress tied to the source paper.

Architecture

warmth-connect-portal/
├── fronted/   # TanStack Start + React 19 + Tailwind CSS + shadcn/ui
├── backend/   # Hono + Drizzle ORM + SQLite + zod
├── desktop/   # Desktop packaging / integration work
├── scripts/   # Integration and validation scripts
└── docs / engineering notes

The AI runtime is integrated as a separate FastClaw service through OpenAI-compatible and streaming APIs.

Technology Stack

Layer	Stack
Frontend	TanStack Start, React 19, Vite, Tailwind CSS v4, shadcn/ui
Backend	Hono, TypeScript, zod, pino
Data	Drizzle ORM, SQLite
AI runtime	FastClaw agents, OpenAI-compatible API, SSE
Remote operations	SSH-based device management
Packaging	Windows-focused desktop engineering loop

Local Development

Frontend

cd fronted
bun install
bun run dev

Backend

cd backend
npm install
cp .env.example .env
npm run db:migrate
npm run dev

The frontend and backend are intentionally runnable independently. Agent features require a configured FastClaw runtime and the corresponding environment variables.

Important API Areas

/api/papers                 Paper library
/api/rag                    RAG conversations
/api/devices                Compute device management
/api/reproduction-records   Reproduction tracking
/api/command                Research command execution
/api/fastclaw               Agent streaming / deployment helpers

What This Project Trains

This project is one of the main long-form projects in my Professional AI Player portfolio. It trains the full path from activity-driven product work to a maintainable AI engineering system:

requirements → product structure → interface → backend contracts → data model → AI integration → validation → packaging

Status

Cheetah Youth Program · Active Development · Full-stack · AI Agent Integration · Research Tooling