Open-Source Projects

Key Features

• Local-first architecture with no cloud dependency for storing personal memory
• CLI for direct interaction and management of stored memories
• MCP server exposing memory tools to AI assistants
• Single-user design focused on privacy and simplicity

Problem Solved

AI assistants lose context between sessions, forcing users to repeat themselves. Recall gives assistants a persistent, local memory store they can read from and write to via MCP, without relying on third-party cloud services.

Key Features

• Auto-routing classifier that picks Solo/Duel/Council execution strategy based on task complexity and stakes
• Multi-model consensus with a judge model flagging disagreement or hallucination
• Real-time cost/analytics dashboard tracking spend and savings vs. a frontier-model baseline
• Fully client-side, no backend, bring-your-own API key via OpenRouter
• Streaming parallel responses across multiple LLMs

Problem Solved

Always routing queries to the most expensive frontier model wastes budget on tasks that don't need it. Maestro dynamically escalates only complex or high-stakes queries to multiple models with built-in answer verification, cutting cost while preserving reliability.

Key Features

• Monitors multiple AI news RSS feeds (The Verge, TechCrunch, MIT, Ars Technica, Wired)
• AI-powered summaries using Perplexity Sonar API
• Automated Telegram delivery with formatted messages
• Daily overview summary consolidating all news into bullet points
• Budget management with configurable article limits (default: 5 articles per run)
• Local storage of all fetched articles and summaries in JSON format
• Time-based filtering (48-hour lookback window)

Problem Solved

Staying updated with AI news requires monitoring multiple sources and spending significant time reading articles. Spectrum automates this process by aggregating RSS feeds, generating concise AI summaries, and delivering them directly to Telegram, enabling users to stay informed with minimal effort while respecting API budget constraints.

Key Features

• Detects bullish/bearish divergences between sentiment and price action
• Real-time analysis via News Scout, Market Analyst, and Judge agents
• Production-ready with Docker, logging, and evaluation tools
• Powered by Google ADK and Gemini 2.5 Flash Lite

Problem Solved

Traders often react emotionally to market news, leading to losses from fear, FOMO, or misinformation. FactFlow eliminates emotional trading by providing transparent, data-driven insights that compare sentiment with actual market performance.

Key Features

• Grid-based scanning algorithm to overcome the Google Places API's 60-result limit
• Interactive map with rectangle drawing tool to define the search area
• Real-time filtering by rating and review count
• Supabase (PostgreSQL + PostGIS) caching to reduce API costs
• Color-coded quality markers for at-a-glance business ranking

Problem Solved

Google Maps search silently caps results at roughly 60 businesses, hiding genuinely excellent options in dense areas. GeoGem systematically scans a region in a grid to surface every business meeting a quality bar that native search would otherwise miss.

Key Features

• Multi-format audio support (MP3, WAV, FLAC, M4A)
• Acoustic fingerprinting using 13 MFCC coefficients
• Euclidean distance-based similarity analysis
• M3U playlist generation compatible with all major players
• Cross-platform compatibility and comprehensive testing

Problem Solved

Users struggle to discover and organize music collections effectively. Auto-DJ intelligently analyzes acoustic properties to generate playlists that maintain musical coherence and flow, providing a personalized music discovery experience.

Key Features

• One-command setup for instant interactive dashboards
• Smart auto-detection of column types with appropriate visualizations
• Multi-dimensional filtering with text search, ranges, and categorical selections
• Auto-generated histograms, bar charts, and time series plots
• Standalone HTML export for offline use
• Optimized for datasets up to 500K+ rows

Problem Solved

Eliminates the need for complex data visualization setup by providing a one-command solution to convert raw data files into professional, interactive dashboards. Addresses the pain point of data analysts who need to quickly explore and share insights without custom dashboard development.

Key Features

• 3-5x faster video conversion with hardware acceleration (NVIDIA, Intel, AMD)
• Intelligent image compression with up to 90% file size reduction
• Professional audio conversion with multiple quality presets
• Modern MVVM architecture with clean separation of concerns
• Real-time progress tracking and error recovery
• 50% reduction in memory consumption

Problem Solved

Users need a fast, efficient way to convert multimedia files without sacrificing quality. FormatFusion provides hardware-accelerated conversion with intelligent optimization, dramatically reducing conversion time and file sizes while maintaining quality.

Key Features

• BERT-based sentiment analysis on financial text
• GPT-based automated report generation summarizing sentiment
• Custom spaCy-based NER model for financial entities (organizations, monetary values, dates)
• Modular pipeline architecture spanning data processing, sentiment analysis, report generation, and NER

Problem Solved

Turning raw financial text and documents into sentiment scores, structured entities, and readable summaries usually requires stitching together multiple manual steps. This pipeline automates the full chain from raw text to actionable report.

Key Features

• Multiple risk assessment models (MVRV, Mayer Multiple, Supply Profitability, NUPL)
• Risk categorization system (Very High, High, Low, Very Low)
• Multi-factor risk assessment framework
• Sophisticated visualizations of risk metrics and price correlations
• Enhanced risk classification accuracy through historical backtesting

Problem Solved

Cryptocurrency markets lack comprehensive risk assessment tools. This project provides institutional investors, algorithmic traders, and researchers with a quantitative framework for understanding Bitcoin market dynamics, reducing lag in identifying trend reversals compared to traditional methods.

Key Features

• Automated data extraction and cleaning using Selenium
• Advanced feature engineering with lagged features and random forest importance
• Multiple ML models optimized for monthly return prediction
• Detailed visualizations of actual vs predicted returns and prices
• 12% improvement in forecast accuracy
• 80% reduction in data preparation time

Problem Solved

Investors need accurate Bitcoin price predictions to manage portfolios effectively. This project provides a robust framework for forecasting that enables investors to predict market trends more accurately and make well-informed decisions regarding crypto assets.

Key Features

• Real-time clipboard monitoring with SQLite persistence
• Support for text, images, and files
• Debounced search, content pinning, and customizable themes
• Global hotkeys for instant content retrieval
• Image processing and text transformations
• Modular, extensible architecture

Problem Solved

Native Windows clipboard only stores the last copied item, causing workflow friction. PastePal transforms the clipboard into an intelligent, persistent content history tool that eliminates productivity losses from lost clipboard data.

Key Features

• Custom dataset upload for industry-specific or proprietary data
• Interactive chat interface with conversational prompts
• Context-aware analysis generating market insights, trends, and predictions
• Chat history management with privacy controls
• Transparent cost monitoring for OpenAI GPT-4 usage

Problem Solved

Businesses need to analyze their proprietary datasets but lack the tools for intuitive, conversational analysis. This project enables users to upload custom datasets and engage in natural language conversations to extract valuable market insights without technical expertise.