Featured Project

Snapfix — AI-Powered VS Code Extension

Open-source VS Code extension that detects Python tracebacks in real time and streams AI-generated fix suggestions. Supports GPT-4, Claude, and local Ollama models interchangeably via LiteLLM.

TypeScriptPythonFastAPILiteLLMOpenAI APIAnthropic APIOllama

View Source Code

TL;DR

Open-source VS Code extension that detects Python tracebacks in real time and streams AI-generated fix suggestions. Supports GPT-4, Claude, and local Ollama models interchangeably via LiteLLM.

The Problem

When debugging Python, developers waste time copying error tracebacks to ChatGPT, losing context between the IDE and the browser. Existing extensions lock you to one AI provider.

My Solution

An open-source VS Code extension that detects Python tracebacks in real time and streams AI-generated explanations and fix suggestions directly into a dedicated side panel — all without leaving the editor.

Technical Architecture

Extension (TypeScript)

VS Code Extension API with real-time terminal output parsing

Regex-based Python traceback detection with context extraction

WebView side panel with streaming markdown rendering

Settings UI for provider selection (no code changes needed)

Backend (Python + FastAPI)

LiteLLM unified interface for multi-provider AI

Structured LLM prompt pipeline for consistent fix quality

Streaming Server-Sent Events for real-time response rendering

Provider health checks with automatic fallback

Multi-Provider AI

OpenAI (GPT-4, GPT-4o) for general error analysis

Anthropic (Claude 3.5) for complex reasoning chains

Ollama (CodeLlama, Mistral) for local/offline support

Zero-config provider swapping — change one setting, same API

Key Metrics

3 AI providers supported simultaneously

Zero code changes to swap between providers

Open-source with full documentation and brand identity

Lessons Learned

The biggest insight was that different AI models excel at different error types. Claude handles complex type errors better, while GPT-4 is faster for simple syntax issues. The multi-provider architecture wasn't just about vendor lock-in — it let me route errors to the best model for each case.

Like what you see?

I can build systems like this for your team. Let's discuss how my skills match your engineering needs.

Let's Talk Email Me Directly

← Previous

Real-Time Attendance Management System

Business Listings Dashboard — Google Maps Lead Scraper