Domain Featured

CodeForge AI

CodeForge solves coding problems through a three-agent pipeline: 1. **Vision Agent (Gemini 2.0 Flash)** — extracts problem statements from uploaded images 2. **Coding Agent (OpenAI o3-mini)** — writes clean, efficient Python with type hints 3. **Execution Agent (OpenAI o3-mini + E2B)** — runs code in a sandbox and explains results Supports image upload, text input, error analysis, and timeout handling.

Python Streamlit

← Back to marketplace

CodeForge AI

Multimodal coding agent by Sharad Khare.

CodeForge AI combines Gemini vision, OpenAI o3-mini coding, and E2B sandbox execution — upload a LeetCode-style problem image or type a description, get Python code, and see it run.

Built and maintained by Sharad Khare — AI strategist, full-stack developer, and creator of practical agent workflows.

What this project does

CodeForge solves coding problems through a three-agent pipeline:

1. Vision Agent (Gemini 2.0 Flash) — extracts problem statements from uploaded images

2. Coding Agent (OpenAI o3-mini) — writes clean, efficient Python with type hints

3. Execution Agent (OpenAI o3-mini + E2B) — runs code in a sandbox and explains results

Supports image upload, text input, error analysis, and timeout handling.

Why Sharad Khare built CodeForge AI

Coding problems often arrive as screenshots — from LeetCode, whiteboards, or PDFs. CodeForge demonstrates how multimodal vision plus reasoning models plus sandbox execution creates a complete solve-and-run workflow in one Streamlit app.

How it works

Image Upload or Text Problem
   │
   ▼
Gemini 2.0 Flash → extract problem statement
   │
   ▼
OpenAI o3-mini → generate Python solution
   │
   ▼
E2B Sandbox → execute code
   │
   ▼
Execution Agent → explain results or errors
   │
   ▼
Streamlit UI with code + execution output

Pipeline stages

| Stage | Model | Role |

|-------|-------|------|

| Vision | gemini-2.0-flash | Extract problem from image |

| Coding | o3-mini | Write Python solution |

| Execution | o3-mini + E2B | Run code and explain output |

Features

Multimodal input: image upload or text description
Gemini vision for problem extraction
o3-mini for code generation and execution explanation
E2B sandbox with timeout and error handling
Modular config, agents, and sandbox modules

Quick start

cd codeforge-ai
pip install -r requirements.txt
streamlit run app.py

Required API keys

OpenAI API key — powers o3-mini coding and execution agents
Gemini API key — powers vision extraction
E2B API key — sandbox code execution

Example inputs

Upload a screenshot of a LeetCode two-sum problem
Type: "Write a function to reverse a linked list with sample I/O"
Upload a photo of a whiteboard coding question

Project structure

codeforge-ai/
├── app.py
├── codeforge/
│   ├── config.py
│   ├── agents.py
│   └── sandbox.py
├── requirements.txt
├── pyproject.toml
└── README.md

Topics

Sharad KhareCodeForge AImultimodal coding agentGemini vision codingo3-mini PythonE2B sandbox executionLeetCode image solverAgno coding agentStreamlit code runnerAI programming assistant.

Who this is for

Developers solving coding problems from screenshots
Students practicing LeetCode-style questions with execution feedback
Engineers exploring multimodal AI + sandbox patterns
Teams prototyping vision-to-code workflows

Use cases

Extract and solve coding problems from images
Generate and execute Python solutions with explanations
Debug sandbox execution errors with AI analysis
Learn Agno + Gemini + E2B integration patterns

Author

Website: sharadkhare.in

License

Loading file tree…

Select a file to preview its source.

bash

cd codeforge-ai
pip install -r requirements.txt
streamlit run app.py