I'm Aaron. I take engineering organizations from zero → one on a fraction of the capital most teams spend — four times as CTO, multiple times as founder, and alongside dozens of founding teams as an advisor and mentor since 2015. Small teams only work when the architecture carries the load.
I evaluate the architecture, AI implementations, and security posture of startups and acquisition targets before capital goes in, and write technical recovery roadmaps for operating companies where delivery is critical. This is the fourth CTO seat; the three before it were venture-backed startups. I have mentored at Techstars since 2015, and I co-founded The Mitten Project to grow the startup base in West Michigan.
System structure mirrors team structure, and the reverse is the useful half: choose the team shape that produces the architecture you want. Communication paths grow at n(n−1)/2, so every hire raises the floor on coordination cost.
Workloads differ by access pattern, not by team. Each specialized component behind a clean port buys query performance and pays for it in operational surface and consistency work — worth it when the access pattern is genuinely different, and not before. Knowing which case you are in is the skill.
Reversibility is a spectrum, and budget should track it. Data models, service boundaries, and security posture sit at the costly end and deserve the deliberation; most of what sits downstream is cheap to rewrite and should move at speed.
My doctoral research captured and extracted situational awareness from social media during mass emergencies. Collective IP created bespoke information retrieval and ML analysis technologies routinely run across 100+ million unstructured records for Fortune 500 biopharma. As a fractional CTO, I ran data science for the consumer division of a Fortune 50 healthcare company and built ML and NLP systems for startups in national air quality, field management, global SEO, health tech, and non-profit. Sopris Health put clinical speech through ML pipelines under HIPAA. Not marketed as AI at the time; the engineering problems were the same ones teams hit today — data quality, drift, evaluation, adaptability, and the cost of inference at scale.
An NLP business intelligence engine that mapped global university tech-transfer IP for Fortune 500 biopharma. Raised $3.5M across seed and Series A. A custom-built 45-node distributed cluster crawled, parsed, and classified more than 100 million unstructured records. Acquired by Wellspring Worldwide.
A HomeKit-enabled smart garden ecosystem launched jointly with Apple, shipped by two core engineers. A fully serverless AWS backend across more than 35 services absorbed global traffic spikes with no downtime and no infrastructure headcount. AWS selected the company to present on stage at re:Invent for the IoT ExpressLink launch. Invited AWS talks for This is My Architecture and IoT fireside chats.
Five months embedded in clinical research at Cedars-Sinai established that a pivot was required; the architecture was rebuilt on custom Deep Learning in two. A unified API over five legacy EHR systems streamed thousands of clinical voice files through ML pipelines under HIPAA.
Data architecture and fractional engineering leadership across early-stage startups and a Fortune 50 enterprise, where I ran data science for the consumer division. A data analytics engine built for one startup client became that company's core platform. Another client raised a $38M Series A/B.
The data collection architecture — based on early Cassandra — for a $2.8M multi-university NSF grant studying public information needs during mass emergencies. It collected more than three billion tweets, the largest in academia at the time, and supplied the data behind 70-plus publications.
On the first team of engineers hired to develop the enterprise agile lifecycle platform, which scaled past 170,000 users and later went public. Technical lead for a seven-person cross-functional team.
Architectures and patterns for polyglot persistence: many purpose-built data stores inside one large-scale application, so a system adapts to specialized storage technologies instead of forcing everything through a single relational database — and a small team can add one without rewriting the application around it.
Access dissertation (PDF)If you are weighing a data-tier decision, sizing an engineering team, or trying to get more product out of the round you already raised — those are my favorite conversations.