Skip to content

PROJECT_INFO.md

Latest commit

 

History

History
53 lines (38 loc) · 1.37 KB

PROJECT_INFO.md

File metadata and controls

53 lines (38 loc) · 1.37 KB

OvaSearch - Project Information

For GSoC 2026 Mentors

This project demonstrates:

  1. Native C++ Performance - Zero Python overhead in runtime
  2. OpenVINO Integration - Full utilization of Intel's inference framework
  3. Multimodal Capabilities - Seamless text and image understanding
  4. Production-Ready Features - Caching, error handling, clean architecture

Quick Demo Commands

# Build from source
mkdir build && cd build
cmake .. && ninja

# Run the application
./ovasearch

# Test queries
"What's in the dog image?"
"Summarize my documents"
"reload" # Show cache performance

Key Technical Achievements

  • 384-dimensional embeddings with cosine similarity search
  • Persistent cache system reducing startup time to milliseconds
  • Sliding window chunking with configurable overlap
  • Automatic document extraction from PDF/DOCX/PPTX
  • Clean C++17 code with RAII patterns

Performance Metrics

  • Document processing: 50-100 docs/second
  • Query latency: <100ms search + 1-3s generation
  • Memory usage: ~500MB base
  • Cache loading: <10ms for thousands of embeddings

Files to Review

  1. main.cpp - Core implementation
  2. CMakeLists.txt - Build configuration
  3. prepare_documents.py - Document preprocessing
  4. pull_model.py - Model download utility

Contact

Created for GSoC 2026 - OpenVINO Organization GitHub: @lagmator22