NK

Portfolio

Nolan Koch

CS & Cybersecurity Developer · Published Researcher

NK

Nolan Koch · CS & Cybersecurity Developer · NeurIPS 2025

From research papers to
production software.

I'm Nolan Koch, a CS & Cybersecurity developer and published AI researcher who turns ambitious ideas into production ready software — full stack apps, automation pipelines, multi-agent LLM systems, and GPU accelerated computing.

Co-author of the Truth-Maintained Memory Agent (NeurIPS 2025). Available for software development projects, research collaborations, and end to end engineering.

View work Read my paper Get in touch
NeurIPS2025 Published
10+Projects shipped
18×CUDA speedup
50k+Records synced
Scroll
NeurIPS 2025Published at ResponsibleFM workshop on reliable foundation models
10+Open source projects across AI, automation, and security
18×GPU speedup over NumPy with CUDA matrix kernels
50k+Customer records synced nightly via production ETL

01 — About

Nolan Koch

CS & Cybersecurity Developer

What I build

I'm a Computer Science & Cybersecurity student at Slippery Rock University and an AI researcher with Algoverse. My co-authored paper on the Truth-Maintained Memory Agent was published at NeurIPS 2025 (ResponsibleFM) — a proactive multi-agent framework for reliable long-context dialogue in LLMs.

I build across the full stack: research grade AI systems, production ETL pipelines, DevSecOps automation, computer vision, GPU kernels, and the web apps that tie it all together. I care about taking ideas from first commit to something that actually runs in the real world.

  • Versatile across Python, C++, TypeScript, SQL, and the full ML/systems/web stack
  • Published NeurIPS 2025 researcher on LLM memory & truth verification
  • Multi-agent systems, vector memory, and reasoning architectures
  • End to end automation: ETL, provisioning, CI/CD security triage
  • CUDA/GPU performance work through full stack web & Flask APIs
  • Off the keyboard: mountain biking, hiking, and gaming

"Ship the idea, not just the slide deck. The best way to understand something is to build it and watch it run."

Full Stack NeurIPS 2025 Published Research Multi-Agent LLMs Python · C++ · TS Machine Learning Automation Cybersecurity GPU / CUDA

— Published research

Peer-reviewed work

NeurIPS 2025 · ResponsibleFM Workshop

Truth-Maintained Memory Agent: Proactive Quality Control for Reliable Long-Context Dialogue

Om Phadke, Jeffrey Guo, Nolan Koch, Andy Xu, Shanduojiao Jiang, Kevin Zhu

Large Language Models are prone to false memory formation during long, multi-turn interactions — incorporating incorrect, irrelevant, or contradictory information. We propose the Truth-Maintained Memory Agent (TMMA), a proactive multi-agent framework that enforces write-time quality control through token-gating, complexity evaluation, and truth-verification across a four-tier hierarchical memory system: Working Memory, Summarized Memory, Archival Memory, and a Flagged Bin for contested content.

Multi-Agent Systems Long-Context Dialogue Memory Management RAG Truth Verification LLMs
Read on OpenReview

02 — Selected work

Case studies

Ten open source builds spanning AI agents, data pipelines, DevSecOps, computer vision, and GPU computing — each designed, implemented, and shipped end to end.

Full archive →
Autonomous support agent with vector memory

01

Autonomous Support Agent

Vector memory AI assistant

RAG-powered agent with vector memory for context-aware responses — built to handle complex, multi-turn interactions with persistent recall.

View project
ETL pipeline syncing customer records

02

ETL Pipeline

50k+ record nightly sync

Production pipeline syncing 50,000+ customer records from legacy MySQL/CRM into HubSpot every night — with validation, error handling, and scheduling.

View project
Zero touch employee provisioning automation

03

Zero Touch Provisioning

Automated employee onboarding

Hands-off employee provisioning workflow — accounts, permissions, and system access triggered automatically from a single HR event.

View project
DevSecOps auto triage pipeline

04

DevSecOps Auto Triage

Security enforced in CI/CD

SecOps pipeline that scans, triages, and enforces security policies across every commit — catching vulnerabilities before they reach production.

View project
Never drop e-commerce router

05

E-commerce Router

High availability routing layer

Fault-tolerant request router for e-commerce traffic — designed to keep storefronts online under load spikes and partial service failures.

View project
Test audit toolkit

06

Test Audit Toolkit

Automated test auditing

Toolkit for auditing test suites — coverage gaps, flaky tests, and quality metrics surfaced automatically for faster CI feedback loops.

View project
Computer vision object detection

07

Object Detection

Real-time YOLOv8 vision

YOLOv8n pipeline for detecting cars and objects in images and live video — optimized for speed with a Flask upload interface and analytics view.

View project
Inventory demand forecasting

08

Inventory Forecasting

Retail demand ML

Time-series forecasting for retail demand using LSTM and Prophet — feature engineering, GPU training, and stock optimization from raw sales history.

View project
CUDA matrix multiplication

09

CUDA Matrix Mult.

18× GPU speedup

CUDA kernels with shared-memory tiling and memory coalescing — multiple variants, benchmarking harness, and Nsight Compute profiling included.

View project
This portfolio website

10

This Portfolio

Vanilla HTML portfolio

Dark editorial portfolio hand-built with vanilla HTML, CSS, and JavaScript — pinned work carousel, research spotlight, and GitHub Pages deploy.

View project

03 — Capabilities

How I deliver

One stack from CUDA kernels to LLM agents to production web — I move comfortably across research, backend, automation, security, and frontend without needing a handoff.

01

AI & ML Systems

NeurIPS-published memory agents, YOLOv8 vision, LSTM/Prophet forecasting, and RL environments — full loops from dataset to trained model to deployed demo.

PyTorch TensorFlow scikit-learn YOLOv8 OpenCV RL / Gym

02

LLMs & Agents

Multi-agent reasoning, vector memory, RAG pipelines, chain-of-thought orchestration, and truth-verification systems — the same patterns behind my published TMMA research.

Multi-Agent RAG Vector DB LangChain Gemini OpenAI API

03

Data & Automation

ETL syncing 50k+ records nightly, zero-touch provisioning, webhook-driven workflows, and LLM-enriched pipelines that run hands-off around the clock.

ETL n8n MySQL Pandas REST APIs Webhooks

04

Backend & APIs

Flask services, REST API design, Dockerized deployments, and routing layers built to handle real traffic, auth patterns, and integration with external platforms.

Python Flask REST Docker JSON SQL

05

Systems & GPU

CUDA kernels with 18× speedups, C++ performance work, CMake builds, and Nsight profiling — low-level engineering when Python isn't fast enough.

CUDA C++ CMake NumPy Parallelism Linux

06

Web, Security & DevOps

Responsive frontends, DevSecOps triage pipelines, CI/CD security gates, and GitHub Actions deploys — shipping code that's fast, secure, and maintainable.

HTML/CSS/JS TypeScript DevSecOps GitHub Actions Cybersecurity Git

Full toolkit

Technologies I work with

Python to C++ to TypeScript — GPU kernels to LLM agents to production ETL. Grouped by practice area, this is the same versatile stack I use across research papers, open source repos, and client-ready builds.

Languages

  • Python
  • C++
  • JavaScript
  • TypeScript
  • SQL
  • HTML / CSS
  • Shell / Bash

AI & Machine Learning

  • PyTorch
  • TensorFlow
  • scikit-learn
  • NumPy / Pandas
  • OpenCV
  • YOLOv8
  • LSTM / Prophet
  • Reinforcement Learning
  • Jupyter

LLMs & Agents

  • Multi-Agent Systems
  • RAG / Vector Search
  • LangChain
  • OpenAI API
  • Google Gemini
  • Chain-of-Thought
  • Memory Architectures
  • Prompt Engineering

Data & Automation

  • ETL Pipelines
  • n8n
  • MySQL
  • HubSpot Integration
  • Webhooks
  • REST APIs
  • JSON / CSV
  • Scheduled Jobs

Backend & APIs

  • Flask
  • FastAPI
  • RESTful Design
  • Docker
  • Microservices
  • Rate Limiting
  • Auth Patterns

Systems & GPU

  • CUDA
  • CMake
  • Nsight Compute
  • Shared Memory Tiling
  • Parallel Computing
  • Linux
  • Performance Profiling

Web & Frontend

  • Responsive HTML/CSS
  • Vanilla JavaScript
  • DOM / IntersectionObserver
  • CSS Grid & Flexbox
  • GitHub Pages
  • Accessibility
  • UI/UX Design

Security & DevOps

  • DevSecOps
  • CI/CD Pipelines
  • GitHub Actions
  • Git / GitHub
  • Security Auditing
  • WAF Concepts
  • Test Automation

Also comfortable with

Python PyTorch LLMs MySQL n8n CUDA Docker Flask JavaScript GitHub Cybersecurity OpenCV TensorFlow Linux TypeScript REST APIs Python PyTorch LLMs MySQL n8n CUDA Docker Flask JavaScript GitHub Cybersecurity OpenCV TensorFlow Linux TypeScript REST APIs

04 — Journey

Where I've shipped

NeurIPS 2025

Truth-Maintained Memory Agent

ResponsibleFM Workshop · Co-author

Published research on proactive quality control for long-context dialogue. TMMA is a multi-agent framework that enforces write-time truth verification to reduce false memory formation in LLMs — using a four-tier memory hierarchy (Working, Summarized, Archival, and Flagged Bin).

Read paper on OpenReview

Present

AI Researcher

Algoverse

Co-authored TMMA — a multi-agent memory framework published at NeurIPS 2025. Also building vector memory agents, ETL systems, and LLM-powered tooling at Algoverse.

Present

CS & Cybersecurity

Slippery Rock University

Coursework and projects spanning AI/ML, cybersecurity, systems programming, and GPU computing — with a focus on shipping working software.

Highlights

10+ Projects Shipped

End to end builds

From vector memory agents and ETL pipelines to DevSecOps automation, YOLOv8 vision, 50k+ record syncs, and an 18× CUDA matrix speedup — each built from first commit to demo.

05 — Contact

Let's build something

Open to full-time roles, contract software development, AI research collaborations, and ambitious side projects. Whether you need a web app, data pipeline, ML integration, or multi-agent system — tell me what you're shipping and let's make it real.

kochnolan376@gmail.com linkedin.com/in/nolankoch github.com/ngk1004

Slippery Rock, Pennsylvania