Work & building
Hackathon builds, company work, research, and personal experiments.
AI pipeline synthesizing trending content from X, YouTube, and Reddit to automate creation for artists and influencers — using Amazon NovaAct and xAI LLM.
Ingests frontend specs via Notion MCP, builds with Vercel V0, tests via Browser Use MCP, and iterates autonomously until every spec passes.
Reduces classroom knowledge disparities through AI-powered peer tutoring using Groq, Perplexity Sonar, and InterSystems IRIS Data Platform.
Natural-language (text, voice, image) interface for government workflows with multi-agent orchestration and XRP Ledger–backed proofs of every AI action.
Multimodal platform using Google Gemini to auto-tag, classify, and severity-score environmental reports, converting citizen inputs into actionable climate intelligence.
Clinical AI using InterSystems IRIS vector search to match patient data against case studies, accelerating pneumonia diagnosis with LLM-powered reasoning chains.
Fine-tuned Qwen-3 on AWS Trainium for real-time GitHub security alert classification. End-to-end pipeline: webhooks → SIEM → WebSockets → automated IAM key rotation.
Prescription scanning, nutrient analysis, and real-time health monitoring with Raspberry Pi hardware + autonomous disaster response simulation with HumeAI emotional intelligence.
29 coordinated uAgents orchestrating Google Veo 3.1, Gemini Flash, Lyria, and TTS to generate fully produced 8-scene cinematic short films from a single text prompt in ~90 seconds.
Designed and implemented a bi-directional adapter between Fetch.ai's uAgents ecosystem and Google's Agent-to-Agent protocol — enabling seamless plug-and-use across both agent networks.
Built integration packages for ASI:One LLM with LangChain, LiteLLM, LlamaIndex, and Skyfire — plus adapters for MCP, LangGraph, and CrewAI agents for Agentverse marketplace discovery.
Enterprise document generation system integrating GPT-4, Claude, and Gemini with custom RAG pipelines via Pinecone — deployed as serverless workflows on AWS Lambda, cutting latency 40%.
ReAct-style orchestration agent retrieving answers from live Discord, Google Docs, and static KB in parallel — with smart escalation routing to human organizers via webhooks when confidence is low.
Domain-specific tools for credit card iframe filling, OTP interception, and date/time selection. Custom tools dramatically outperform pure LLM approaches for repetitive, high-stakes form interactions.
Fine-tuned Qwen-3 on AWS Trainium for narrow classification tasks. Finding: well-curated training data on a small model can match frontier models at a fraction of inference cost.
Stateful LangGraph agent with Square Payments integration serving 100+ attendees. The hard part: keeping conversation context coherent as people rejoined mid-order hours later.