Copilot-Style Inline Suggestions
Ghost text completions appear directly in the editor as you type Azure SDK code — accept with Tab, dismiss with Escape.
Project Overview
Doppelganger is a full-stack developer tool consisting of a VS Code extension, an Express backend API, and a React-based marketing website — built in an intense 30-hour hackathon sprint.
The extension detects when a developer is working with Azure SDKs (storage, cosmos, identity, keyvault, service-bus, etc.) and provides inline ghost text suggestions powered by Groq's Llama model. A RAG pipeline using ChromaDB ensures suggestions are grounded in real Azure documentation, drastically reducing hallucinations.
The backend orchestrates the entire pipeline — context analysis, vector retrieval with SDK-specific filtering, prompt construction, and response cleanup. A feedback endpoint with MongoDB persistence enables continuous improvement of suggestion quality. The extension also features automatic Azure import injection, intent detection from comments, and a multi-layer caching system (session + Redis) that reduces API calls by 80%+.
While working with Azure SDKs, developers often need to repeatedly go through documentation and copy commands manually — slowing down the development process. Our solution simplifies this workflow by providing AI-powered suggestions directly inside VS Code, helping developers use Azure SDKs without constantly switching to documentation.
🎥 Video Explanation
🎯 Project Purpose
Built during the Doppelganger 30-Hour Build Sprint hackathon organized by OpenPools — where Team Eklavya secured 1st Rank among 33 teams. The extension eliminates AI hallucinations in Azure SDK code suggestions by using context-injected Azure documentation retrieval (RAG) to provide accurate, grounded inline completions for JS/TS/C# developers.
🛍️ Project Checkouts (Key Highlights)
Azure-Aware Context Detection
Automatically detects Azure SDK usage in JS/TS/C# (storage, cosmos, identity, keyvault, service-bus) and tailors suggestions accordingly.
RAG-Powered Documentation Grounding
ChromaDB + embedding service retrieves relevant Azure docs to ground every suggestion in real documentation — no hallucinations.
Auto Import & Quick Fixes
Automatically injects missing Azure SDK imports and provides quick fix code actions for common patterns.
Feedback & Continuous Improvement
Built-in feedback endpoint with MongoDB persistence allows rating suggestions for continuous model improvement.
Mock Mode for Local Development
One boolean switch enables full local development without cloud dependencies — perfect for offline coding.
🧩 Technology Stack
Frontend
React (Marketing Website + VS Code Webview Panel)
Backend
Node.js, Express.js (API Service + RAG Pipeline)
Cloud & Infrastructure
MongoDB (Feedback Storage), ChromaDB (Vector Store)
Core Architecture
TypeScript (VS Code Extension Host), Groq Llama LLM
🧠 Challenges & Solutions
Problem: AI code assistants often hallucinate Azure SDK APIs that don't exist.
Solution: Implemented a RAG pipeline using ChromaDB with embedded Azure documentation, ensuring every suggestion is grounded in real, verified SDK references.
Problem: YouTube Shorts-style SPA navigation in VS Code can cause content scripts to lose context.
Solution: Intercepted VS Code editor events, used MutationObserver patterns, and implemented a singleton architecture to ensure the extension reinitializes correctly on context changes.
Problem: High API call volume during rapid typing creates latency and cost issues.
Solution: Built a multi-layer caching system (session + Redis) achieving 80%+ API call reduction, combined with a debounced typing watcher for intelligent trigger timing.
Problem: Developers need to quickly switch between mock and production environments during development.
Solution: Designed a single-boolean configuration switch that seamlessly toggles between mock fallback behavior and full cloud backend — zero config changes needed.
📂 Project Structure
azure-ai-code-extension/extension/src — VS Code extension host, inline provider, code watcher, Azure detector
azure-ai-code-extension/backend/src — Express API, RAG service, LLM integration, feedback controller
azure-ai-code-extension/extension/webview — React app rendered in VS Code panel
azure-ai-code-extension/extension/frontend — React marketing/demo website
azure-ai-code-extension/shared — Shared constants and types across packages
🏆 Hackathon Achievement
🏆 1st Place winner
Doppelganger 30-Hour Build Sprint — organized by OpenPools.in
👥 Team Members
Dhruvesh ShyaraFrontend Development & Project Workflow Coordination
Priy MavaniBackend Developer (API Development & Deployment)
Arjun DivraniyaExtension Logic Pipeline & Integration
Mayank DudhatraRAG System, Research & Feedback System
🙏 Special Thanks
CodingGita & Patel Neel Maheshkumar — for continuous support, mentorship, and guidance throughout the hackathon journey.
✨ Key Learnings
- The power of team collaboration under pressure
- Building real developer tools in a limited time
- Turning an idea into a working product within 30 hours
⭐ Why This Project Stands Out
- 🏆 1st Place Winner: Secured 1st Rank among 33 teams at the Doppelganger 30-Hour Build Sprint by OpenPools.
- Published on VS Code Marketplace: Live extension available for immediate download and use.
- Zero Hallucination Architecture: RAG pipeline eliminates the #1 problem with AI code assistants.
- Security-First: Always suggests DefaultAzureCredential over connection strings.
- Intent Detection: Detects developer intent from comments (e.g., '// I need to upload to blob' → generates full function).
- 30-Hour Sprint: Entire project — extension, backend API, RAG pipeline, and marketing website — built in just 30 hours.