Dynamic Reasoning LLC · Mahwah, NJ
Software that holds up
in production.
I’m Nicholas Korcynski. I build applied AI, multi-vendor commerce, and end-to-end web systems for teams that need to put real software in front of real customers.
Selected work
2025—present
Who’s Brew
Multi-vendor specialty coffee marketplace · whosbrewshop.com
Who’s Brew set out to fix the fragmented specialty-coffee market: every roaster runs their own storefront, every checkout is separate, every customer pays shipping a dozen different ways. The brief was to build a single marketplace where independent vendors share a unified cart, a single checkout, and a transparent shipping flow — without losing the per-vendor autonomy that makes specialty coffee specialty.
I designed and built the whole system. Next.js on AWS. MongoDB with a deliberately split data layer for customer and staff concerns. Redis for sessions, rate limiting, and idempotency. Stripe Connect for per-vendor payouts with server-side price re-resolution at every step. Shippo for centralized fulfillment. A Cerebras-served Llama model driving natural-language search and personalized recommendations. Five user roles with department-scoped permissions, an immutable store-credit ledger, and a passport feature that tracks every country a customer has sourced from. Real money moves through it every day.
What I do
Four things, done well. Each one drawn from work that’s already in production.
Production web & commerce
Multi-vendor marketplaces, payments, fulfillment, role-based admin. The kind of system that needs to be right under load and right in an audit.
Applied LLM features
Designing and shipping LLM features that work in production — search, recommendations, content generation — and that don’t blow up the cost model.
Computer vision
Local-first CV pipelines for documents, whiteboards, and workflow automation. OpenCV plus hardware-accelerated inference where it earns its place.
ML model development & deployment
Training, exporting, and serving real models — usually as small inference APIs that deploy anywhere.
Also recent
2025
SceneIt
Gemini-driven movie discovery on AWS. Decoupled React frontend and App Runner backend with Google OAuth and a multi-axis preference model.
2025
OneNote Whiteboard Scanner
Windows desktop app that digitizes whiteboards locally with OpenCV plus ONNX DirectML and OpenVINO. Ships the result straight into OneNote via Microsoft Graph.
2025
AI Chess Engine
AlphaZero-inspired ResNet with Monte Carlo Tree Search, trained on Lichess games and served as a lightweight ONNX REST API on CPU.
About
I’m a software engineer at The Home Depot. Dynamic Reasoning is the practice I run alongside that role.
Production React on homedepot.com by day. Applied AI and full-stack consulting outside of it. AI-engineering certifications from IBM, Microsoft, and NVIDIA, and a body of shipped systems — including Who’s Brew above — long enough to know which patterns survive contact with production.
The work I take on tends to share a few traits: it’s customer-facing, it has to be right the first time, and there isn’t a five-person team to throw at it.
More about me →