OneAI — AI Orchestration Platform

An intelligent AI orchestration system that routes user prompts to the best AI model, decomposes complex tasks, and supports multi-model comparison with an LLM judge.

---

What It Does

1. User enters a prompt in a ChatGPT-like interface
2. Planner decides whether to split the prompt into subtasks
3. Classifier labels each task (coding, research, writing, analysis, general)
4. Router maps categories to AI providers (Groq, Gemini, OpenRouter)
5. Enhancer improves the prompt before sending it to the model
6. Tasks run in parallel with asyncio.gather()
7. Responses are combined and returned to the user

Bonus — Compare Mode: Send the same prompt to all 3 models, then let a Judge LLM pick the best response.

---

Tech Stack

Layer	Technology
Frontend	Next.js, TypeScript, Tailwind
Backend	FastAPI
Database	PostgreSQL 16
AI APIs	OpenRouter, Gemini, Groq
Deploy	Docker Compose

---

Folder Structure

OneAI/
├── backend/                    # FastAPI Python backend
│   ├── app/
│   │   ├── main.py             # App entry point
│   │   ├── config.py           # Environment settings
│   │   ├── database.py         # DB connection
│   │   ├── api/
│   │   │   └── routes.py       # HTTP endpoints
│   │   ├── models/
│   │   │   └── db_models.py    # SQLAlchemy tables
│   │   ├── schemas/
│   │   │   └── api.py          # Request/response shapes
│   │   ├── providers/          # AI API adapters
│   │   │   ├── base.py         # Abstract base class
│   │   │   ├── openrouter.py
│   │   │   ├── gemini.py
│   │   │   ├── groq.py
│   │   │   └── factory.py      # Provider lookup
│   │   └── modules/            # Orchestration logic
│   │       ├── planner.py      # Task decomposition
│   │       ├── classifier.py   # Category labeling
│   │       ├── router.py       # Provider mapping
│   │       ├── enhancer.py     # Prompt improvement
│   │       ├── judge.py        # Best response picker
│   │       └── orchestrator.py # Main coordinator
│   ├── requirements.txt
│   └── Dockerfile
├── frontend/                   # Next.js React frontend
│   ├── app/
│   │   ├── layout.tsx
│   │   ├── page.tsx            # Main chat page
│   │   └── globals.css
│   ├── components/
│   │   ├── Sidebar.tsx
│   │   ├── MessageList.tsx
│   │   ├── ChatInput.tsx
│   │   ├── ComparePanel.tsx
│   │   └── LoadingIndicator.tsx
│   └── lib/
│       ├── api.ts              # Backend HTTP client
│       └── types.ts            # TypeScript interfaces
├── database/
│   └── schema.sql              # PostgreSQL tables
├── docs/
│   ├── MODULES.md              # How each module works
│   ├── API_EXAMPLES.md         # curl examples
│   └── MOCKUPS.md              # UI wireframes
├── docker-compose.yml
├── .env.example
└── README.md

---

Quick Start

Step 1: Configure Environment

# Copy the example env file and add your API keys
cp .env.example .env

Step 2: Start Database + Backend (Docker)

docker compose up -d

This starts:

• PostgreSQL on port 5432 (auto-runs database/schema.sql)
• FastAPI backend on port 8000

Step 3: Start Frontend

cd frontend
cp .env.local.example .env.local
npm install
npm run dev

Open http://localhost:3000

---

Local Development (Without Docker)

Backend

# Start PostgreSQL yourself, then:
cd backend
python -m venv venv

# Windows
venv\Scripts\activate


pip install -r requirements.txt

# Run the schema manually against your Postgres:
# psql -U oneai -d oneai -f ../database/schema.sql

uvicorn app.main:app --reload --port 8000

Frontend

cd frontend
npm install
npm run dev

---

Example Usage

Simple question:

"What is machine learning?" → Single task → classified as research → routed to Gemini

Coding request:

"Write a binary search in Python" → Single task → classified as coding → routed to Groq

Complex request:

"Explain blockchain AND write a Python implementation" → Planner splits into 2 tasks → run in parallel → combined response

Compare mode:

Toggle "Multi-Model Comparison Mode" → all 3 models respond → click "Choose Best Response" → Judge picks winner

---