Trees in a quiet valley, with mist over the hills.
回首向来萧瑟处,也无风雨也无晴。 — 苏轼

Agentic AI Systems / Security & Privacy

Hi, I'm Takku.

Building agents.
Thinking about trust.

I’m a Founding Engineer at Noetic, building privacy-focused AI tools for reading and reflection.

My focus is on agentic AI systems, especially the security and privacy challenges that arise when agents interact with tools, data, and people.

Experience

Previously, I worked as a Machine Learning Engineer at Cognitiv, through September 2026. I earned my M.S. in Computer Science at Northeastern University in December 2025.

Selected projects

2 projects
VeilGate concept: an agent accesses resources through a privacy boundary.

Security & privacy

VeilGate

Privacy-Preserving AI Gateway

A privacy-preserving resource access gateway for AI agents. Exploring the boundary between agent capabilities and access to private resources.

  • Agent systems
  • Privacy

Earlier work

BJMR social graph: connected people form three communities linked by shared conversations.

Business Job Market Rumors

BJMR

An anonymous discussion platform for the business-school academic job market. Combines NLP-assisted content moderation with context-based authentication and device management.

  • Full-stack
  • NLP moderation
  • Authentication
Eloquence pronunciation coaching: one speech bubble says Eloquence; the other says EL-uh-kwuhns, with stress on the first syllable.

Eloquence

A mobile-first AI speaking coach for interviews, negotiations, and difficult conversations. Combines speech input and coach personas with post-conversation scoring and actionable feedback.

  • Conversational AI
  • Speech
  • Coaching
Game recommendation concept: a query retrieves review documents to inform a recommendation.

RAG-Based Game Recommendation System

A natural-language game recommendation API grounded in Steam reviews. Uses LangChain and FAISS to retrieve relevant reviews and generate recommendations with supporting evidence.

  • RAG
  • Vector search
  • Recommendations
Async4Ai task graph: a workflow branches into parallel tasks and joins at the next stage.

Async4Ai

A lightweight Go framework for asynchronous scheduling of multi-stage deep learning tasks. Supports continuous data-processing workflows, with a web API for updating and querying tasks.

  • Go
  • Async scheduling
  • ML infrastructure