# Matt Hodges

> Matt Hodges is a political technologist and founder of Ilium Strategies, a consulting practice helping Democrats build, scale, and innovate with technology and AI. Former Director of Engineering for Biden 2020. Featured in The New York Times, Wired, The Washington Post, and The Economist.

## About

I am a **Political Technologist** and founder of
Ilium Strategies, where I provide
senior technical and strategic advising to Democrats building, scaling,
and innovating with technology. Drawing on over fifteen years of
experience at the intersection of Democratic politics and software
engineering, I work with leaders, campaigns, mission-driven
organizations, donors, and investors. My focus areas include AI strategy
and adoption, product and engineering vision, go-to-market technical
strategy, donor advising and technical due diligence, technical
leadership for core ecosystem infrastructure, and political tech market
navigation. I’m particularly interested in developing sustainable,
scalable models for long-term impact in the political tech sector.

I sit on the **Advisory Board** of Higher
Ground Labs, an accelerator that invests in early-stage political
tech companies. In this capacity, I contribute strategic guidance to the
Higher Ground Labs leadership team, and provide both political and
technical advice to investment cohort companies.

I was the **Executive Director** of Zinc
Labs, a political innovation lab at **Zinc Collective** that
houses the organization’s strategic work in Democratic technology and
pilot programs. Through a network of campaigns, committees, vendors,
donors, and subject matter experts, Zinc Labs incubates and invests in
innovative new programs across the ecosystem. I prioritized investing in
technology and people to enable collaborative experimentation, technical
execution, and hypothesis testing.

I was on the **Advisory Board** of
Quiller, an Artificial Intelligence
copilot that helps Democrats draft and deploy high quality, effective
fundraising content. In this capacity, I contributed product,
technology, and political guidance to the Quiller leadership team.
Quiller was acquired in 2025.

I was the **Director of Engineering** for
President Joe Biden on his 2020
presidential campaign. I joined the campaign in July of 2019, and served
in that role through the duration of the primary and general election
cycles. I led a team of software, infrastructure, and security
engineers, which was charged with delivering innovative new campaign
tech for headquarters and states programs. We shipped dozens of internal
and voter-facing products, including an in-house peer-to-peer texter, a
relational organizing app, a real-time debate transcription service, a
document and media ingestion and indexing library, a live donation
tracking web service, a distributed soft-reporting platform, the
JoeBiden.com web property, hundreds of ETL pipelines, and a cross-region
data warehouse.

I was a **Senior Software Engineer** on **The Groundwork** platform,
which served as the application and infrastructure system that powered
the 2016 presidential campaign of Secretary
Hillary Clinton. Beginning in 2015, I was the primary engineer
for multiple campaign-enabling web services for event signups,
single-click donations, and peer-to-peer fundraising.

I was an **Infrastructure Engineer** at
Datadog where I architected and
implemented the internal platform for Incident Response management. This
tool eventually joined the suite of observability offerings available to
customers. My team also implemented the organization’s first templated
deployment pipeline for the organization’s global transition to
Kubernetes.

I was a **Senior Mobile Engineer** at
LISNR, where I developed the product’s
first iOS SDK for the ultrasonic data transport protocol. In 2014 the
team joined the TechStars accelerator. This toolchain was originally
designed to allow musicians to embed inaudbile data streams into their
media. Over time, it became a location-aware second screen immersion
platform for music festivals, professional sporting events, and feature
films. It now serves as a payment protocol in partnership with VISA.

I was a **Senior Mobile Engineer** and **Scrum Master** at
Cardinal Solutions, with the team that
built Vantiv Accept - a mobile credit card processing application.
Envisioned as a competitor to Square, the application was deployed to
retail and grocery point-of-sale systems across the country.

I am a **Named Inventor** on an Augmented
Reality Patent Application (US20140092241A1) for a “device and
method for scanning of books and other items for order and inventory
control”. We created a mobile application that scanned a shelf of books
in real-time as a user viewed it through the screen. If it identified
any items out of order, it overlayed the user’s view of the item with a
red X to indicate a misplacement, and then povided directional guidance
of where it should be moved.

I have been an **Election Night Reporter** for
Decision Desk HQ. I gathered live
results from New Mexico and Ohio Boards of Elections offices to report
real-time data as they were delivered from precincts. These results were
featured on Vox and BuzzFeed election night coverage.

I have been a frequent **Open Source Contributor** to
OpenElections, which is creating the
first free, comprehensive, standardized, linked set of election results
data for the United States.

[Full bio, media, and appearances](https://matthodges.com/index.md)

## Posts

- [Thank you.](https://matthodges.com/posts/2020-11-10-thank-you/index.md): Working on the Biden Campaign was the opportunity of a lifetime.
- [Exploring the Lambda Calculus with Python](https://matthodges.com/posts/2022-07-17-exploring-lambda-calculus-python/index.md): Some basic ideas of the Lambda Calculus, and how to use it to implement a computation system in Python
- [Building a Neural Network From Scratch with NumPy](https://matthodges.com/posts/2022-08-06-neural-network-from-scratch-python-numpy/index.md): Let's walk through creating a Neural Network from scratch, only using NumPy, and build a model that classifies the MNIST handwritten digit data set.
- [BIDEN: Binary Inference Dictionaries for Electoral NLP](https://matthodges.com/posts/2023-10-01-BIDEN-binary-inference-dictionaries-for-electoral-nlp/index.md): A compression-based binary classification technique that is fast at both training and inference on common CPU hardware in Python
- [Language Models on the AI Executive Order](https://matthodges.com/posts/2023-11-01-language-models-ai-executive-order/index.md): Deploying text embeddings and Llama2 to answer questions about the Executive Order on Artificial Intelligence issued by President Biden
- [Amused Entirely To Death](https://matthodges.com/posts/2024-07-20-amused-entirely-to-death/index.md): What if the imminent danger to our democracy isn't just looming authoritarianism, but also our own insatiable appetite for distraction? Last week’s Republican National Convention provided a stark reminder of this unsettling threat.
- [How Did Skannerz Work?](https://matthodges.com/posts/2024-07-23-how-did-skannerz-work/index.md): Only 90s kids are now adults who dig through patent filings to understand how their toys worked.
- [Austin, Texas: Hot or Not?](https://matthodges.com/posts/2024-07-30-austin-hot-or-not/index.md): This summer hasn't felt as hot as last summer. I decided to find out if that was actually true.
- [Spline Versus PCHIP](https://matthodges.com/posts/2024-08-08-spline-pchip/index.md): Sometimes you want a smooth curve. Sometimes you want a shape-preserving curve.
- [CSVs Are Kinda Bad. DSVs Are Kinda Good.](https://matthodges.com/posts/2024-08-12-csv-bad-dsv-good/index.md): I would simply use delimiter characters to delimit data.
- [ActBlue Isn't Selling Your Data](https://matthodges.com/posts/2024-08-25-actblue-isnt-selling-your-data/index.md): The campaign you donated to might be sharing your data.
- [I Cracked And Modded the DNC Bracelet](https://matthodges.com/posts/2024-08-28-cracking-modding-dnc-bracelet/index.md): The Democratic National Convention lit up the United Center with bracelets. I wanted to light mine up at home.
- [Part 2: Austin Texas, Hot or Not?](https://matthodges.com/posts/2024-12-26-austin-hot-or-not-part-2/index.md): Confronting temperature vibes with graphs, because that tactic always works.
- [GPT Finally Jumps Out of the System](https://matthodges.com/posts/2025-04-21-openai-o4-mini-high-mu-puzzle/index.md): OpenAI’s o4-mini-high Model Solves the MU Puzzle and Demonstrates Why
- [Using LLM Embeddings to Normalize User Data](https://matthodges.com/posts/2025-08-02-language-model-embeddings-campaign-donors/index.md): How cosine similarity in embedding space can power high-quality normalization.
- [I Donated to Sherrod Brown via ChatGPT](https://matthodges.com/posts/2025-08-20-chatgpt-agent-political-donation/index.md): I let the LLM agent send money through ActBlue
- [Music to Break Models By](https://matthodges.com/posts/2025-08-26-music-to-break-models-by/index.md): Gödelian limits of prompt-safe AI
- [ChatGPT Got Me Reading Plato](https://matthodges.com/posts/2025-09-23-chatgpt-plato-republic/index.md): Notes, Prompts, and Noble Lies: A New Way to Jump Into Old Philosophy
- [Scraping the Federal Reserve with VisiData](https://matthodges.com/posts/2025-09-30-visidata/index.md): Big banks, small commands: a one-liner to wrangle the Fed's bank list
- [Tracking Five Years Of Health Data](https://matthodges.com/posts/2025-10-12-five-years-of-health-data/index.md): A thousand rides, a stack of apps, and actual motivation to get in shape
- [The Privacy Theater of Hashed PII](https://matthodges.com/posts/2025-10-19-privacy-theater-pii-phone-numbers/index.md): A 2020 MacBook Air can hash every North American phone number in four hours
- [Dad](https://matthodges.com/posts/2025-11-02-dad/index.md): I love my daughter so much. I love my wife so much.
- [Claude in a Game Theory Tournament](https://matthodges.com/posts/2025-12-14-claude-axelrod-prisoners-dilemma/index.md): How Claude Code ran a competitive Iterated Prisoner's Dilemma strategy
- [My New Agentic Campaign Finance Sidekick](https://matthodges.com/posts/2025-12-19-ai-agent-fec/index.md): Building a campaign finance analyzer using Claude’s Agent Skills
- [Campaigns Are Knowledge Workers and the Tools Just Caught Up](https://matthodges.com/posts/2026-01-07-ai-agents-campaigns/index.md): Agentic AI can change campaign operations. The organizations that outlast November need to lead.
- [Quick DataViz With Claude Code](https://matthodges.com/posts/2026-01-20-quick-dataviz-claude-code/index.md): General-purpose agents for everyday data tasks
- [The Feed Framed the Last Election and AI Answers Will Frame the Next](https://matthodges.com/posts/2026-03-20-democrats-answer-layer/index.md): Democrats need to start building for the AI answer layer.
- [Surviving Hallucinations: AI Fault Tolerance for Campaigns](https://matthodges.com/posts/2026-04-10-ai-politics-fault-tolerance/index.md): Building AI workflows that campaigns can trust under pressure
- [A Magazine With One Subscriber](https://matthodges.com/posts/2026-07-12-personal-magazine-claude-codex/index.md): How three AI agents turn the articles I save into a real printed magazine.
- [Bongard Problems](https://matthodges.com/posts/2026-08-19-bongard-problems/index.md): There's a blurry line between pattern matching and reasoning.