Projects

A complete project archive spanning distributed AI systems, developer tools, data-intensive applications, and full-stack products.

AI / ML Systems

4 projects

Conductor: Distributed Multi-Agent Workflow RuntimeDistributed AI SystemsApr 2026 - Jun 2026A Kubernetes-native Go runtime for crash-resumable multi-agent workflows, distributed scheduling, and event-sourced observability.

Tech stack

Go, Kubernetes, Kafka, Event Sourcing, DAG Orchestration, Multi-Agent Systems

Description

A Kubernetes-native Go runtime for crash-resumable, multi-agent LLM workflows with durable execution, distributed scheduling, and event-sourced observability.

Achievements

  • Added per-step checkpoints so long-running workflows can resume after crashes without duplicating tool calls or model spend.
  • Scheduled DAG-based workflows across a scalable Kafka-backed worker pool with per-task retries.
  • Captured every model and tool call through event-sourced tracing, audit logs, and real-time workflow monitoring.
Engram: Cross-Session Memory MCP for AI Coding AgentsAgent Memory and EvaluationJan 2026 - Mar 2026A Python MCP memory harness for carrying decisions, project facts, and preferences across AI coding sessions while detecting stale context.

Tech stack

Python, MCP, Event-Driven Architecture, Agent Memory, BM25, Evaluation

Description

A Python MCP memory harness that carries decisions, project facts, and preferences across AI coding sessions while detecting stale or conflicting context.

Achievements

  • Reduced repeated context tokens by approximately 40% through cross-session state handoffs.
  • Consolidated raw sessions into durable short- and long-term memory with stale-memory cleanup and conflict detection.
  • Validated retrieval reliability with Recall@5, MRR, and traces for ranking regressions and injected context.

Metrics

  • ~40% fewer repeated context tokens
  • Recall@5 and MRR evaluation
FigBrain: Figma Multi-Agent PluginMulti-Agent PluginApr 2025 - May 2025An MCP-powered Figma agent that uses a LangGraph supervisor and ReAct tool loops to turn natural-language prompts into FigJam components.

Tech stack

Python, TypeScript, LangGraph, MCP, Figma API, PostgreSQL

Description

An MCP-powered Figma agent built for the Microsoft AI Hackathon, using a LangGraph supervisor and ReAct tool loops to turn natural-language prompts into FigJam components.

Achievements

  • Converted natural-language prompts into FigJam components at approximately 85% first-try accuracy.
  • Reduced API costs by approximately 28% and lowered latency with PostgreSQL-backed memory, caching, and model routing.

Metrics

  • ~85% first-try accuracy
  • ~28% lower API costs
AI Recognition in Edge Computing & Cloud NativeAI/ML & Edge ComputingMay 2023 - July 2023A KubeEdge and TensorFlow vegetable-recognition system with Django and MQTT management plus a Kubernetes CI/CD pipeline.

Tech stack

KubeEdge, TensorFlow, CNN, Django, MQTT, Kubernetes, CI/CD, Edge Computing, Cloud Native

Description

Optimized edge computing AI recognition using KubeEdge, streamlining local computations and decision-making. Devised a Convolutional Neural Networks model in TensorFlow for vegetable recognition, achieving 98.6% accuracy. Built a centralized management system using Django with MQTT, enhancing live oversight in multiple edge modules. Fostered system resilience and prompt updates by integrating a Kubernetes orchestrated CI/CD pipeline.