I refined a method to pool GPUs from different vendors and generations into a single inference backend of roughly 20 GB over Vulkan. FreeToken, a bleeding-edge MoE inference server, fronts the cluster and handles the elastic inference, running Mixture-of-Experts models with far more total parameters than the cluster's VRAM would normally allow. My own scripts handle KV caching and system-prompt compaction, which is what makes the setup harness-capable.
The same rig trains as well as it serves, having refined Qwen SLMs and diffusion image models. All this is hosted alongside my personal web assets, like this website. I voluntarily subject myself to the maintenance of production-grade systems on every level of the tech stack to ensure that I survive the Big Tech apocalypse.
",
+ "stack": ["FreeToken", "ComfyUI", "PEFT / TRL", "Docker", "Cloudflare", "Reverse Proxy"],
+ "image": "photos/deepdives/homelab-dashboard.png",
+ "caption": "Cluster dashboard during a long inference run."
+ },
+ "capstone": {
+ "title": "Subatomic Particle Tracing and Anomaly Control",
+ "tagline": "My Master's Capstone: Building a Diffusion Cloud Chamber and using stereo reconstruction to classify Muons and Alpha/Beta particles.",
+ "body": "
My ongoing Data Science Capstone Project combines existing utilities for cheap quantum demonstrations and real world measuring techniques to create effective sensing of subatomic phenomena without breaking the bank or into the IAEA
",
+ "stack": ["Ego"]
+ }
+}
diff --git a/src/static/json/pages.json b/src/static/json/pages.json
index 7fdbcd3..b4a8ff8 100755
--- a/src/static/json/pages.json
+++ b/src/static/json/pages.json
@@ -7,7 +7,7 @@
},
"status": {
"template": "status.html",
- "title":"Andrew Simonson - Status Page",
+ "title": "Andrew Simonson - Status Page",
"description": "Status page for my services",
"canonical": "/status"
},
@@ -25,7 +25,7 @@
},
"duck": {
"template": "duck.html",
- "title":"You've been ducked!",
+ "title": "You've been ducked!",
"description": "Face it, you've been ducked",
"canonical": "/duck"
},
@@ -38,31 +38,31 @@
"ecolab": {
"title": "Data Scientist / Data Engineer — Ecolab",
"date": "January 2024 – Present",
- "content": "Data Scientist / Data Engineer Ecolab · Saint Paul, MN
• Primary model developer of RushReady from inception through multimillion-dollar product commercialization • Engineered near-real-time data platform with Delta Live Tables and automated MLOps lifecycle across QSR brands • Built explainable anomaly detection with XGBoost/SHAP attribution and brand-configurable recommendations, increasing speed of service by 20% • Represented Ecolab as liaison in Microsoft's 100-member AI accelerator program",
+ "content": "
Ecolab · Saint Paul, MN
Primary model developer of RushReady from inception through multimillion-dollar product commercialization
Engineered near-real-time data platform with Delta Live Tables and automated MLOps lifecycle across QSR brands
Built explainable anomaly detection with XGBoost/SHAP attribution and brand-configurable recommendations, increasing speed of service by 20%
Represented Ecolab as liaison in Microsoft's 100-member AI accelerator program
",
"classes": "experience"
},
"rit_ms": {
"title": "M.S. Data Science — RIT",
"date": "Dec 2026 (expected)",
- "content": "Master of Science in Data Science Rochester Institute of Technology · Rochester, NY
• Focus on probability theory, statistical learning, and Bayesian methods • Capstone: Fall 2026",
+ "content": "
Rochester Institute of Technology · Rochester, NY
Focus on probability theory, statistical learning, and Bayesian methods
• Independently built reactive chemistry analysis Flask app encoding adiabatic correction and exotherm-detection logic across 9 test types • Architected object-model abstraction, decoupling test-type parsers from report generation and replacing 4 legacy VBA tools • Product became the interdepartmental data standard, saving >1,020 hours (~0.65 FTE) annually",
+ "content": "
Dow Chemical · Freeport, TX
Independently built reactive chemistry analysis Flask app encoding adiabatic correction and exotherm-detection logic across 9 test types
Architected object-model abstraction, decoupling test-type parsers from report generation and replacing 4 legacy VBA tools
Product became the interdepartmental data standard, saving >1,020 hours (~0.65 FTE) annually
",
"classes": "experience"
},
"rit_bs": {
"title": "B.S. Computer Science & Data Science — RIT",
"date": "Dec 2024",
- "content": "B.S. in Computer Science · B.S. in Data Science (Dual Cluster) Rochester Institute of Technology · Rochester, NY
• GPA 3.63 · Dean's List • Minor in International Relations • MicroMasters in Data Science – UC San Diego (edX)",
+ "content": "
Rochester Institute of Technology · Rochester, NY
GPA 3.63 · Dean's List
Minor in International Relations
MicroMasters in Data Science – UC San Diego (edX)
",
"classes": "education"
},
"csh": {
"title": "E-Board Member & Presenter — CSH",
"date": "Aug 2021 – Present",
- "content": "E-Board Member & Presenter Computer Science House · Rochester, NY
• Presented seminars on web scraping, analytics, and GIS • Built and maintained web services for 80+ members",
+ "content": "
Computer Science House · Rochester, NY
Presented seminars on web scraping, analytics, and GIS
Built and maintained web services for 80+ members
",
"classes": "experience technical"
}
}
diff --git a/src/static/json/projects.json b/src/static/json/projects.json
index d547e96..6002baa 100755
--- a/src/static/json/projects.json
+++ b/src/static/json/projects.json
@@ -25,28 +25,28 @@
]
]
},
- "PsyCom - Physical Combinatorics": {
+ "PhysCom - Physical Combinatorics": {
"status": "WIP",
"classes": "programming",
"content": "Experimental innovation engine operating on physical attributes and limitations of proven existing technologies with further AI review."
},
"Antietam-Conococheague Watershed Monitoring": {
"status": "complete",
- "classes": "geospacial",
+ "classes": "geospatial",
"bgi": "watershedTemps.png",
- "content": "Live geospacial analysis of Maryland's Antietam and Conococheague sub-watersheds, monitoring water quality and temperatures through the summer months for governmental environment health review boards."
+ "content": "Live geospatial analysis of Maryland's Antietam and Conococheague sub-watersheds, monitoring water quality and temperatures through the summer months for governmental environment health review boards."
},
"Automotive Brand Valuation Analysis": {
"status": "complete",
"classes": "programming",
"bgi": "automotiveBrandAnalysis.png",
- "content": "Brand valuation analysis of the used car market, measuring value decay by mileage to extrapolate qualities such as percieved reliability and persistent value of luxury features."
+ "content": "Brand valuation analysis of the used car market, measuring value decay by mileage to extrapolate qualities such as perceived reliability and persistent value of luxury features."
},
"RIT Hotspots": {
"status": "incomplete",
- "classes": "pinned geospacial programming",
+ "classes": "pinned geospatial programming",
"bgi": "hotspotsrit.png",
- "content": "Live crowd migration map using RIT occupancy data. It seems RIT didn't like me exposing their surveilance state but since they didn't want to talk to me about it they instead changed the service response schema a few times. When that didn't stop me they just shut down the whole service. Nerds.",
+ "content": "Live crowd migration map using RIT occupancy data. It seems RIT didn't like me exposing their surveillance state but since they didn't want to talk to me about it they instead changed the service response schema a few times. When that didn't stop me they just shut down the whole service. Nerds.",
"links": [
["github", "https://github.com/asimonson1125/hotspotsrit", "git repo"]
]
@@ -55,14 +55,14 @@
"status": "complete",
"classes": "pinned programming",
"bgi": "calorimeterAnalysis.png",
- "content": "An analytical toolkit designed for reactive chemistry analysis, especially calorimetry. Works include automatic analysis, alerting unusual and dangerous results derived from a wide range of testing envrionments and equipment",
+ "content": "An analytical toolkit designed for reactive chemistry analysis, especially calorimetry. Works include automatic analysis, alerting unusual and dangerous results derived from a wide range of testing environments and equipment",
"links": []
},
"Geography of Alternative Energy": {
"status": "complete",
- "classes": "pinned geospacial",
+ "classes": "pinned geospatial",
"bgi": "energyGeography.png",
- "content": "An ArcGIS geospacial analysis comparing the difference in effectiveness of wind, solar, and geothermal energy across the continental 48 United States.",
+ "content": "An ArcGIS geospatial analysis comparing the difference in effectiveness of wind, solar, and geothermal energy across the continental 48 United States.",
"links": [
[
"globe",
diff --git a/src/static/photos/books/hailmary.jpg b/src/static/photos/books/hailmary.jpg
new file mode 100644
index 0000000..ef5a4c9
Binary files /dev/null and b/src/static/photos/books/hailmary.jpg differ
diff --git a/src/static/photos/books/sherlockholmes.jpg b/src/static/photos/books/sherlockholmes.jpg
new file mode 100644
index 0000000..1a5295d
Binary files /dev/null and b/src/static/photos/books/sherlockholmes.jpg differ
diff --git a/src/static/photos/books/themartian.jpg b/src/static/photos/books/themartian.jpg
new file mode 100644
index 0000000..a32fd10
Binary files /dev/null and b/src/static/photos/books/themartian.jpg differ
diff --git a/src/static/photos/extradimensional.avif b/src/static/photos/extradimensional.avif
new file mode 100644
index 0000000..7911737
Binary files /dev/null and b/src/static/photos/extradimensional.avif differ
diff --git a/src/static/photos/omnidimensional.avif b/src/static/photos/omnidimensional.avif
new file mode 100644
index 0000000..5211022
Binary files /dev/null and b/src/static/photos/omnidimensional.avif differ
diff --git a/src/static/photos/test.png b/src/static/photos/test.png
new file mode 100644
index 0000000..370381e
Binary files /dev/null and b/src/static/photos/test.png differ
diff --git a/src/templates/about.html b/src/templates/about.html
index c1d7078..89d8c49 100644
--- a/src/templates/about.html
+++ b/src/templates/about.html
@@ -1,211 +1,132 @@
{% block content %}
-
-
-
-
-
About
-
Bio, timeline & credentials
-
+
+
+
About
+
+ I'm Andrew Simonson, a data scientist at Ecolab and a graduate data
+ science student at Rochester Institute of Technology, where I recently
+ finished a B.S. in Computer Science with a minor in international
+ relations and a focus on probability theory.
+
+
+
+
+
+ At work I build the models and the pipelines behind them. At home I run
+ the servers. In between, I get bored and put things on this website,
+ self-built and self-hosted as a sandbox for
+ webdev experiments. This is what unprofessional development looks like.
+
+
+ There is also a résumé for paper-minded people.
+
-
-
-
-
-
- I'm Andrew Simonson,
- a Data Scientist at Ecolab and a graduate Data
- Science student at
- Rochester Institute of Technology, having
- recently completed the Computer Science BS program
- (international relations minor) with a focus on probability
- theory.
-
-
- I get bored and throw random stuff on this website.
- This is what unprofessional development looks like.
-
-
-
- I also have a
- resume
- for unexplained reasons.
-
+ Data scientist at Ecolab and a graduate student at RIT.
+ I build end-to-end decision machines: from data pipelines that process scale, to epistemic models which understand, to user interfaces that convey.
+
+ Primary model developer on RushReady at Ecolab: explainable anomaly
+ detection with XGBoost and SHAP, plus brand-configurable recommendations
+ for quick-service restaurants, all running on a near-real-time Delta
+ Live Tables platform.
+
+
+
+
At school
+
+ M.S. in Data Science at Rochester Institute of Technology, focused on
+ probability theory, statistical learning, and Bayesian methods.
+ Capstone in fall 2026.
+
+
+
+
At home
+
+ Running a two-GPU homelab
+ that serves Mixture-of-Experts models to coding agents over local
+ endpoints, and using the same rig to fine-tune small models.
+
+
+
+
On the shelf
+
+
+ {% for title in var.books.selection[:5] %}
+ {% set book = var.books.books[title] %}
+
+ Research, tooling, and experiments. Sometimes I forget to add things to this list, message me for missing details.
+
+
+
+
+
+
Highlights
-
-
-
-
-
-
Highlights
-
- {% from 'partials/feature_project.html' import feature_project %}
- {{ feature_project(
- 'homelab',
- 'AI Homelab',
- 'Self-hosted GPU cluster serving MoE models beyond their rated VRAM, with a full LoRA fine-tuning pipeline on top',
- '
I\'ve refined a method to pool GPUs from different vendors and generations — an RTX 5070 and a Radeon RX 6600 XT — into a single ~20GB inference backend over Vulkan, so neither card\'s limits bottleneck the other. FreeToken, a bleeding-edge MoE inference server, fronts the cluster and handles the actual elastic inference, running Mixture-of-Experts models with far more total parameters than the cluster\'s VRAM would normally allow. My custom scripts resolve KV caching and system prompt compaction that make the setup harness-capable, so agentic tools like Claude Code can hit it over Anthropic- and OpenAI-compatible endpoints exactly like they would a paid API. GPU layer placement alone took generation from ~150s a call to ~88ms/token.
The same rig trains as well as it serves. I built a complete LoRA fine-tuning pipeline on it — base model, training config, and adapter. Used for fine-tuning Qwen3-1.7B on patent data, and training a Stable Diffusion LoRA set on curated, deduplicated image data.
',
- ['RTX 5070', 'RX 6600 XT', 'Ollama', 'FreeToken', 'ComfyUI', 'PEFT / TRL'],
- 'photos/deepdives/homelab-dashboard.png'
- ) }}
-
- {{ feature_project(
- 'siteinfra',
- 'Self-Hosted Web Tooling & Maintenance',
- 'I designed, deployed, and still actively operate this site\'s stack — the app, the edge, and its own health monitoring',
- '
This portfolio runs on infrastructure I designed, deployed, and still operate myself — the Flask app and its container, the Cloudflare edge and reverse proxy in front of it, DNS and tunneling, and the uptime monitor watching all of it (live at /status). None of that is a managed host\'s job here; it\'s mine.
What actually proves it\'s production-capable is the maintenance loop, not the launch. I put it through a real adversarial review periodically — most recently a live pass with Claude Code against both the deployed site and the source — and fix what it finds. The latest pass closed two real issues same-day: a database-query pattern on an internal endpoint that could be hammered into needless load, and a self-built redirect that could momentarily hand a browser plaintext HTTP instead of HTTPS. Nothing catastrophic came up — no injection, no auth bypass, no data exposure — which is exactly the point of doing this on a schedule instead of waiting for something to break.
}})
+ +
+ {% endfor %}
+ - -
 }})
 }})
+ Andrew Simonson
+
+
-
+ onclick="toggleMenu()"
+ >Menu
+
 }})
- + +
+ {% endfor %}
+  }})
-  }})
-  }})
-  }})
-  }})
-  }})
+  }})
+ {% if caption %}
+  }})
{% else %}
-
+
{% endif %}
![{{ i[0] }}]({{ url_for('static', filename=src) }})
- {{ i[2] }}
-
- {% endif %}
+