Skip to content

About

AI/ML engineer building agentic systems that are verifiable, secure, and governable, from production RAG agents to research on formal verification and privacy-preserving machine learning.

Education

Integrated BS / MS in Computer Science

University of Minnesota, Twin Cities

GPA 3.8 / 4.0

May 2023 - May 2028

  • Artificial Intelligence
  • Agentic Systems
  • Cryptography
  • Deep Learning
  • MLOps on AWS
  • Parallel Computing
  • Algorithms & Data Structures
  • Data Science
  • Event-Driven Architecture
  • Cyber Security
  • Cloud Computing

What I work on

  • Agentic Systems
  • Retrieval-Augmented Generation
  • Machine Learning
  • AI Engineering
  • MLOps on AWS
  • Federated & Distributed Learning
  • Automated Reasoning
  • Computer Vision
  • Cryptography

In short

I build agentic systems. At Canaan Group I own the delivery of an AI sales platform where agents extract structured data from conversations, propose follow-ups, and flag anomalies against a semantic knowledge base.

The other half of my work is governance: making those systems verifiable, secure, and auditable rather than trusted on faith. That means privacy-preserving federated learning, zero-knowledge proofs over model updates, and formal verification of agent actions before they run.

Shipping agents is the easy part. Proving one did not leak, drift, or act outside policy is the part that decides whether it belongs anywhere serious.

I am an Integrated BS/MS Computer Science student at the University of Minnesota, and I spend most weekends on CTF competitions, mostly reverse engineering and cryptography.

Experience

  1. June 2026 - Present

    Agentic Systems Architect

    Canaan Group

    Richmond, Canada

    • Own end-to-end delivery of a conversation-extraction AI sales app (React Native, Cloudflare R2, LangGraph), cutting rep data entry from 3 hours per week to sub-30-second reviews, capturing 5x more missing customer fields in the CRM, and replacing the Salesforce workflow for $40K/year in license savings.
    • Led the company's agentic transformation by building a semantic knowledge base RAG pipeline on AWS Bedrock that integrates with the CRM, letting AI agents propose follow-ups, surface revenue opportunities, and flag shipment and customs anomalies at 90%+ extraction accuracy while cutting missed follow-ups by 40%.
    • Operate forward-deployed across sales, engineering, marketing, and the executive team to scope, pilot, and productize the platform as a freemium App Store SaaS projecting $75K first-year ARR.

  2. Jan 2026 - Present

    AI Researcher

    UMN Distributed Systems Lab

    Minneapolis, MN · NSF-funded ($1.1M)

    • Contribute to a $1.1M NSF-funded project on privacy-preserving federated learning, helping build tools that evaluate whether AI systems can protect sensitive data while staying robust under attack.
    • Developed modular attack and defense plugins that let researchers test new privacy and security scenarios without rewriting the training system, reducing experiment setup from 3 days to 1 hour.

  3. June 2025 - October 2025

    AI Engineer Intern

    HeadstarterAI

    New York, NY

    • Built an internal AI automation platform: Python microservices and a Next.js admin console that ingest documents, run Langchain RAG over internal knowledge bases, and extract structured entities using OCR and Document AI.
    • Implemented a verification pipeline integrated with CRM REST APIs to sync validated records, targeting 95-98% automation while maintaining 100% verified entries through human-in-the-loop review, cutting 4 hours of manual work per week.

  4. Sep 2025 - Nov 2025

    ML-RSA UROP Research

    UMN Distributed Machine Learning Systems Lab

    Minneapolis, MN · $2K UROP grant

    • Earned a $2K Undergraduate Research Award to study approaches for RSA semiprime factorization under Dr. Ali Anwar, building 4 model architectures (LSTM, Transformer, CNN-RNN, GAN) across 4 dataset scales.
    • Engineered mathematical features using ECPP and GNFS-inspired methods, achieving 63% exact-match accuracy and a 548x improvement over random chance.
    • Reached 91.6% accuracy within 4-bit tolerance on 210K semiprimes, the first documented application of Transformer attention to RSA factorization.