Senior Full-Stack Developer · Blockchain Core Contributor · AI Infrastructure Engineer
13 years building software. Professional since 16. Terminal-native on Linux. Chemistry, physics, and HPC research that I actually use.
Flavormancer — Predicts how a molecule tastes and smells from its chemical structure alone. 190 trained heads across four sensory modalities — 6 taste, 167 aroma descriptors, 5 mouthfeel/chemesthesis, 12 Tox21 safety screens — over ~8,900 molecules, with a 178-dimensional flavor-profile vector behind substitution search and a 3D flavor-space map on real MW × logP × TPSA axes.
Two things I care about more than the model:
- It runs entirely on hardware you own. No cloud call, no data leaving the building. Same thesis as the rest of my AI work, applied to food & beverage R&D.
- It tells you when it doesn't know. Every head publishes its own calibrated decision threshold and its measured out-of-fold precision, because AUROC is blind to class imbalance and will happily report 0.98 for a head that's right 10% of the time. Heads that can't clear a 50% precision floor ship marked indicative rather than quietly dressed up as confident. Training data is public-domain and open-government only — commercially clean by construction, not by promise.
Python · FastAPI · RDKit · scikit-learn · UMAP · Parquet · ONNX · MCP server · Claude skill
Live and viewable now.
Private AI infrastructure — On-premises LLM deployments for small businesses, and other entities. Fully offline inference using Ollama/vLLM, RAG pipelines with vector databases, GPU-accelerated on commodity server hardware. No data leaves the building.
Sunroom CRM — Dual-stack CRM built to demonstrate full technical breadth. .NET Core/C#/SQL Server backend + Laravel/PostgreSQL backend, with frontends in Angular, React, Vue, and Blazor across six repos: Angular · React · Vue · Laravel API · Blazor · .NET API
A consolidated
sunroom-crmhub repo is planned to tie the six together and hold the shared web shell and assets.
countylocal.net — Hyperlocal community platform with wildcard subdomain architecture designed to scale county-by-county. Laravel 12 · React 19 · PostgreSQL · Typesense · Redis · Laravel Reverb · Meade County Local Repo
| Project | What it is |
|---|---|
| Flavormancer | Structure → taste / aroma / mouthfeel / safety prediction. 190 RandomForest heads on Morgan fingerprints + physicochemical descriptors, per-head threshold calibration, and a generalization audit that catches heads which memorize instead of learn. On-prem, commercially clean data |
| Ravencoin / KawPow ecosystem | Core contributor maintaining consensus-critical C/C++ codebase, Python libraries, ElectrumX servers, and cross-platform Electrum wallet builds |
| Satori blockchain | Full cryptocurrency fork through testnet — genesis block, airdrop to ~29K holders, stratum mining proxy, wallet builds (AppImage + Windows) |
| SatoriP2P / Satori-lite | libp2p-based decentralized P2P layer for the Satori AI prediction engine — Chronos + XGBoost models across a distributed peer network |
| python-evrmorelib / python-ravencoinlib / python-satorilib | Python library enhancements from Ravencoin to Evrmore and Satori blockchains: UTXO management, transaction construction, Base58Check encoding |
| Enterprise web apps | ASP.NET Core APIs, Laravel TALL stack applications, Node.js services, Unity multiplayer game systems, UWP kiosk applications |
Backend C#/.NET Core · Laravel/PHP · Python · FastAPI · Node.js · C/C++
Frontend React · Vue.js · TypeScript · Tailwind CSS · Livewire · Inertia.js
Data PostgreSQL · MySQL · MariaDB · SQL Server · Redis · Typesense · MongoDB · ChromaDB · pgvector
Sci/ML RDKit · scikit-learn · NumPy · pandas · UMAP · XGBoost · AmberMD · Parquet · ONNX
AI/ML Claude Code · Model Context Protocol · Ollama · vLLM · RAG pipelines · CUDA inference
Infra Linux · systemd · Docker · Kubernetes · AWS · NGINX · Apache · Cloudflare · Proxmox · GitHub Actions
Testing pytest · Playwright · CI gating on lint + unit + headless E2E
Hardware Dell PowerEdge servers · GPU rigs · Board-level electronics repair
Western Kentucky University — dual major B.S. in ACS-certified Chemistry and Physics, minors in Astronomy and Mathematics (magna cum laude).
Graduate coursework toward an M.S. in ACS Chemistry, specializing in HPCC computational research (degree not conferred). Student researcher running AmberMD molecular dynamics simulations on high-performance computing clusters — modeling rhodopsin protein conformations and SO₂/water interfaces.
That's where the obsession with GPU computing started, and it's why Flavormancer exists: cheminformatics is where my two careers stop being separate things.
Co-Owner of Echelon Technology Solutions, LLC — IT, web development, AI consulting, and more in Meade County, Kentucky.
More about how I work
I build from the terminal. Claude Code is my primary AI development workflow. I run my own inference servers on Dell PowerEdge hardware. I mine cryptocurrency on rigs I built and maintain. When a client's PS5 needs a new HDMI port, I do the board-level swap with liquid metal reapplication.
I've been writing code since I was 12 and charging for it since I was 16. I've touched everything from molecular dynamics on supercomputer clusters to WordPress plugins to blockchain consensus algorithms. The thread connecting all of it is that I'd rather understand the whole stack than depend on someone else's abstraction. There isn't a tech challenge that I am not willing to give my best efforts, I am a strong believer in the idea that we learn our best through solving new problems!
That instinct shows up in how I treat models, too. A metric that flatters the system is worse than no metric, so I build the test that can embarrass me — firing every trained head across the entire corpus to find the ones that only recognize their own training set, and publishing the precision number right next to the one that looks good. Something honest about its limits beats something impressive that quietly isn't.