# Selected work: situations, decisions, and what came of them.

A small selection of roles and engagements, framed around the decisions involved rather than a list of responsibilities.

## Bringing AI-assisted takeoffs to the glass industry

2023 – Present · AI / computer vision · Vertical SaaS · Product strategy

**Situation:** Glazier Software serves glass and glazing contractors who priced jobs by manually measuring plans, a slow and error-prone process with no computer-vision-assisted estimating tool anywhere in the vertical.

**Role:** Chief Technology Officer, owning product strategy and technical architecture, and partnering directly with sales and executive leadership on the roadmap.

**Approach:** Directed development of a proprietary computer vision model to automate takeoffs, the glass industry's first tool of its kind, while building and scaling the engineering organization delivering the broader project management, invoicing, and operations platform contractors rely on as their system of record.

**Outcome:** The AI takeoff tool became a differentiator in sales conversations, with direct involvement contributing to roughly 75% of closed deals and about $750K in new revenue.

## Cutting infrastructure costs 90% on a high-volume ad data pipeline

2023 · Data pipelines · Cost optimization · Cloud architecture

**Situation:** Leaf Advertising's data processing pipeline handled millions of advertising transactions but had grown costly to run and slow to process.

**Role:** Engineering Lead over a 6-person mobile, backend, and frontend team building on Angular, Node.js, and Google Cloud Platform.

**Approach:** Rewrote the pipeline architecture end to end, targeting the specific bottlenecks driving both compute cost and processing latency.

**Outcome:** Reduced infrastructure costs by 90% and improved processing performance by 10x.

## Rebuilding the data platform for a $1B ecommerce business

2019 – 2023 · Data platforms · Team building · Ecommerce

**Situation:** Cresco Labs needed a reliable data platform to support a 3,000-person, roughly $1B-revenue company, after this engagement began as a greenfield ecommerce build.

**Role:** Director - Data Engineering & BI. Started as principal engineer on the ecommerce platform, then grew into managing 4 data engineers, 1 data scientist, and 7 BI resources including 1 manager.

**Approach:** Overhauled the data warehouse architecture using Docker, ECS, Airflow, and Python for reliable pipelines, and rolled out a new quarterly planning methodology to increase visibility and execution accountability across the team.

**Outcome:** Built an ecommerce backend responsible for $125M in monthly revenue and created a data engineering team that attracted talent from across the company.

## Pricing every vacation rental fast enough to sort by price

2008 – 2016 · Distributed systems · Pricing engines · Travel marketplaces

**Situation:** At Expedia, a vacation rental's nightly rate isn't a stored number. It's the output of a rate engine each property manager configures: seasonality, length of stay, occupancy, lead time before arrival, taxes, fees, and discounts, normalized across seven or more tables in separate systems. Quoting one property meant a database round trip. Quoting a page of search results, or sorting that page by price, was out of reach, so travelers compared properties on a base rate that wasn't what they'd actually pay.

**Role:** Software Architect, then Engineering Manager, then Senior Software Engineering Manager. Owner of the rate quoting platform across its life, and a named inventor on the patent it produced.

**Approach:** Two tiers. Aggregators fan a request out across properties; quoters each own a partition of the rate data in denormalized in-memory caches, compute the rate components in parallel, and hand them back for reassembly into a single quote. Requests carry a verbosity level, so a booking page gets the full line-item breakdown while a search page gets hundreds of bare totals at once. Caches refresh in place, so owners can change their rates without taking quoting down.

**Outcome:** At least 750,000 quotes per second, accurate to the penny. That made sort-by-price and price filtering work against live search results instead of an approximation. The architecture became U.S. patent US20170132695A1, since abandoned.

Reference: [US20170132695A1](https://patents.google.com/patent/US20170132695A1/en)

---

Dave Brown · CTO & Executive Advisor · New York, NY · [hi@davecto.com](mailto:hi@davecto.com)

Site guide for agents: [llms.txt](https://davecto.com/llms.txt) · Sitemap: [sitemap.xml](https://davecto.com/sitemap.xml) · Any page is available as Markdown by sending `Accept: text/markdown` or adding `.md` to its URL.