Autonomous Multi-Agent Systems & Swarm Intelligence
Autonomous multi-agent orchestration, decentralized swarm intelligence, self-improving agent factories, Reinforcement Learning with Verifiable Rewards (RLVR / GRPO), tool calling, Model Context Protocol (MCP), and multi-agent safety oversight.
Hackathon Projects & Wins:
- Eephor: Oversight Agent for Multi-Agent Systems: Specialized oversight LLMs detecting deceptive agents in multi-agent environments (MoltBook/Ethos Academy), estimating hidden
SOUL.mdprofiles, LoRA/PEFT via Unsloth, and RLVR via Group Relative Policy Optimization (GRPO) in OpenEnv. Published moltbook-oversight-llama31-1b and moltbook-oversight-llama-3.2-3b-instruct models on Hugging Face. (Reinforcement Learning Hackathon by Cerebral Valley 2026) - ¡alacard!: The Self-Improving Cookbook for AI Agents: Open-source instruction layer and agentic factory that composes, executes, traces, and autonomously repairs Jupyter notebooks using Google ADK, Daytona sandboxes, W&B Weave State-Action-Reward tracing, and Vertex AI / W&B Inference code patch generation. Finalist (top 6 projects) at WeaveHacks 2: Self-Improving Agents Hackathon 2025.
- Stigmergy ODEA (Open Distributed Edge Agents): Decentralized multi-agent swarm intelligence for tactical edge operations and first responders using on-device Google Gemma 3B multimodal models, Android Nearby Connections P2P mesh network, dynamic leader election, and prompt injection defense. (Google Gemma 3n Impact Challenge on Kaggle 2025)
- AI-Q Research Assistant with Universal Deep Research (UDR): Two-level agentic system (LangGraph orchestrator + UDR strategy-as-code dynamic Python executor) synthesizing NVIDIA NIM microservices (Nemotron-Super-49B, Llama-3.3-70B, NeMo Retriever), Docling ingestion, AWS EKS + Karpenter GPU auto-scaling, and CopilotKit streaming UI. (AWS & NVIDIA Agentic AI Unleashed Hackathon 2025)
- BarackOllama & BarackOllamaCLI: Agentic personalized resume, GitHub portfolio, and cover letter customization suite powered by Google Agent Development Kit (ADK), Gemini, Snowflake, and CrewAI Enterprise. (Llama Lounge Snowflake Agentic Hackathon 2026)
- Public Health Application: Public health surveillance system built with FastAPI, LangGraph intelligent agents (
PublicHealthDashboardAgent), multi-LLM support (GPT-4 / Claude), and FastMCP servers over SSE. (AI Expo SCSP Hackathon by Booz Allen Hamilton & SCSP 2025)
Core Competencies: Multi-Agent Orchestration (Google ADK, LangGraph, CrewAI), Reinforcement Learning (RLVR / GRPO, Unsloth, TRL), Decentralized Swarms & P2P Edge Networks, Agent Evaluation, Tracing & Oversight, Tool Use & Model Context Protocol (FastMCP).