Project Portfolio
Nine projects across five technology waves — from LLM APIs and RAG systems to autonomous agent harnesses.
FinEdge
completed LLM APIsFintech investment intelligence
Built a financial intelligence platform that uses LLMs to analyze market data, generate investment insights, and automate portfolio reporting. Integrated Anthropic APIs with PostgreSQL for persistent knowledge storage.
45K LOC · Python + Anthropic + PostgreSQL
- Automated daily market analysis across 500+ equities
- Reduced investment research time by 80% with AI-generated briefs
- Built custom LangGraph pipelines for multi-step financial reasoning
SuperCRM
completed Agent FrameworksAgentic social CRM
Developed an AI-powered CRM that uses Claude agents to enrich contacts, draft personalized outreach, and manage social relationships. Pioneered early agent-framework patterns with structured tool use.
35.7K LOC · Next.js + Claude AI Agents
- Managed 10K+ contacts with AI-enriched profiles
- Automated outreach generation with context-aware personalization
- Integrated multi-channel social data into unified contact graph
KnowledgeGraph
completed RAG SystemsStrategic intelligence & benchmarks
Created a knowledge management platform with interactive D3.js visualizations for mapping research domains, benchmarking AI models, and surfacing strategic insights from large document corpora.
81.7K LOC · Next.js + D3.js + React
- Visualized 2,000+ interconnected research nodes with D3.js force graphs
- Built interactive benchmark dashboards for AI model comparison
- Processed 500+ research papers into structured knowledge graphs
AgentKit
completed MCP ProtocolVisual agent testing platform
Designed a desktop application for visually building, testing, and debugging AI agent workflows. Used Tauri for native performance with a React frontend and Pydantic-validated agent schemas.
473 tests · React + TypeScript + Tauri
- Created visual drag-and-drop agent workflow builder
- Achieved 473 tests with comprehensive agent behavior coverage
- Implemented MCP protocol support for standardized agent communication
Canvas OS
completed Agent OrchestrationAnalytics & data visualization
Built an interactive canvas-based analytics platform with React Flow for node-based data pipelines and Zustand for performant state management. Focused on real-time data visualization and exploration.
28.4K LOC · React + TypeScript
- Rendered complex data pipelines with 100+ connected nodes
- Built real-time analytics dashboards with sub-100ms updates
- Designed extensible plugin system for custom visualization widgets
DeepResearch
completed Autonomous AgentsGrounded AI research assistant
Developed an autonomous research agent that breaks complex questions into sub-queries, searches multiple sources, synthesizes findings, and generates cited research reports with LangGraph orchestration.
36 blueprints · Python + LangGraph + Vue
- Generated 36 research blueprints with full source citations
- Implemented multi-hop reasoning across diverse knowledge sources
- Built Vue-based interface for interactive research exploration
BrowseAI
completed AI InterfacesAI browsing companion
Created a browser extension that brings AI assistance directly into the browsing experience. Supports multiple LLM providers (Anthropic, OpenAI, Gemini) for page summarization, Q&A, and content extraction.
21.9K LOC · Browser extension + multi-provider LLM
- Shipped cross-browser extension with 3 LLM provider integrations
- Built intelligent page context extraction for grounded responses
- Implemented streaming responses with real-time markdown rendering
TerminalOS
completed AI Form FactorsCLI workflow automation
Built a terminal-native AI assistant that orchestrates CLI workflows across multiple LLM providers. Supports Anthropic, OpenAI, Gemini, and local Ollama models with unified command interface.
15K LOC · Anthropic + OpenAI + Gemini + Ollama
- Unified 4 LLM providers under a single CLI interface
- Automated complex multi-step terminal workflows
- Implemented local-first model support via Ollama integration
Stagent
active The Harness LayerYour business, run by AI
Shipped a local-first AI business operating system that orchestrates agents across your entire company — with 5-runtime execution, smart routing, governance, cost controls, chat UI, multi-channel delivery (Slack/Telegram), 25 workflow patterns, 44+ database tables, and a living book documenting the journey.
111K LOC · 426 tests · 116/128 features shipped
- Shipped 116 features across 37 operator surfaces in 28 days — open source, local-first
- 5-runtime architecture with smart router, 52+ agent profiles, and 25 workflow patterns
- Full chat system, heartbeat scheduling, multi-channel delivery, and human-in-the-loop governance