---

AI Recipe Generator (Multi-Reasoning + Voice)

An end-to-end AI-powered recipe generation system supporting advanced LLM reasoning strategies — Chain-of-Thought (CoT), ReAct, and Tree-of-Thought (ToT) — along with voice-based input.

Built using FastAPI (backend), React (frontend), and Gemini LLM, this project demonstrates modern LLM orchestration, reasoning pipelines, structured outputs, and clean result management.

Features

Reasoning Modes

• Chain-of-Thought (CoT) – step-by-step logical reasoning
• ReAct – structured reasoning with action-oriented responses
• Tree-of-Thought (ToT) – multi-sample reasoning with scoring and best-answer selection
• Voice Input – generate recipes by speaking instead of typing

---

Dashboard

• View responses for each reasoning mode
• Timestamped, structured outputs
• Unified result view across CoT, ReAct, and ToT

---

Results Management

• All final outputs are stored in a single tracked directory:

backend/results/

• No unnecessary nested folders for ToT
• Clean, readable JSON outputs for inspection and debugging

---

Project Structure

AI-recipe-generator/
│
├── backend/
│ ├── main.py
│ ├── models/
│ │ ├── cot.py
│ │ ├── react.py
│ │ └── tot.py
│ │
│ ├── gemini/
│ │ └── client.py
│ │
│ ├── utils/
│ │ ├── results_manager.py
│ │ └── memory_manager.py
│ │
│ ├── results/ # MAIN results folder (tracked in Git)
│ │
│ └── requirements.txt
│
├── frontend/
│ ├── src/
│ │ ├── components/
│ │ │ └── Dashboard.jsx
│ │ └── App.jsx
│ │
│ └── package.json
│
├── .env
├── .gitignore
└── README.md

---

Setup Instructions

1.Backend Setup

cd backend
conda activate transformer-2   # or your preferred environment
pip install -r requirements.txt

Create a .env file:

GEMINI_API_KEY=your_api_key_here
USE_MOCK_GEMINI=false

Run the backend:

uvicorn main:app --reload

Backend will run on:

http://127.0.0.1:8000

---

2.Frontend Setup

cd frontend
npm install
npm run dev

Frontend will run on:

http://localhost:5173

---

API Endpoints

Method	Endpoint	Description
POST	/cot	Chain-of-Thought reasoning
POST	/react	ReAct reasoning
POST	/tot	Tree-of-Thought reasoning
POST	/voice	Voice-based recipe generation

---

Technical Highlights

• Gemini Python SDK with prompt-enforced JSON outputs
• Safe JSON parsing with graceful fallback handling
• No unsupported SDK schema usage
• Modular, extensible reasoning architecture
• Production-safe FastAPI backend design
• Mock mode support for quota-free testing

---

Notes

• backend/results/ is intentionally version-controlled
• backend/utils/results/ is ignored to avoid local clutter
• Mock mode enables testing without consuming Gemini API quota

---

👨‍💻 Author

Aryaman Jain AI / ML / LLM Engineering