Generative AI Development
Agentic systems, multi-agent frameworks, and AI tools I've built
AgentSpeak: Agentic Simulation Framework
A 'Market Simulation' platform capable of modeling 100k consumer personas to test policies, pricing, and messaging in realistic scenarios. Evaluated agent alignment, consistency, and emergent behaviors through statistical analysis to ensure reliable, realistic, and interpretable outcomes.
Scale: 100k consumer persona simulations
Use Cases: Policy testing, pricing strategies, marketing messaging
Focus: Agent alignment, consistency, emergent behavior analysis
Investi: Autonomous Investing System
An autonomous investing system using multi-agent collaboration (Researcher, Analyst, Trader) to execute logic-based trading strategies without human intervention. Features Alpaca APIs for execution, PostgreSQL with RAG for memory search and state persistence, and LangSmith tracing for observable workflows.
Architecture: Multi-agent (Researcher, Analyst, Trader)
Execution: Alpaca APIs for trading
Memory: PostgreSQL with RAG for state persistence
Observability: LangSmith tracing for workflow monitoring
Explicit Agent: Minimalist Agent Framework
A minimalist agent framework that exposes all agent logic, tool calls, and state management, enabling full transparency and user control over agent behavior. Features modular tools and state management for specific context persistence across complex multi-step tasks.
Philosophy: Full transparency and user control
Features: Exposed agent logic, tool calls, state management
Design: Modular tools for multi-step task persistence
Multi-Agent Systems Research
Open-source frameworks for multi-robot collaboration and human-robot interaction
Collaborative Gym: Multi-Robot Interaction Framework
An open-source framework for modeling and simulating multi-robot interaction in cooperative tasks. Built with OpenAI Gym, this environment enables researchers to develop and test multi-agent reinforcement learning algorithms for collaborative robotics scenarios.
Research Focus: Multi-robot cooperation and coordination
Institution: MAGICS Lab, Northeastern University
Reaching Task: Human-Robot Collaboration
A multi-agent reinforcement learning framework for human-robot collaboration that achieved 95% accuracy in collaborative reaching tasks. This research explores how robots can learn to anticipate and coordinate with human partners in shared workspaces.
Achievement: 95% accuracy in human-robot collaboration task
Institution: MAGICS Lab, Northeastern University