ML that increases profits
Localize and classify objects that will destroy your AV
Autonomous vehicles need to avoid collisions, so the machine learning models that drive them need massive volumes of labeled data. Getting that data can be prohibitively expensive, so we sped up the process with our own ML to localize and classify objects like helicopters. Additionally, a separate model detected frames with no target objects, allowing us to quickly skip labeling those.
Key outcome
Sequences with target objects now usually need only verification and take seconds instead of minutes. Skipping empty frames saved time as well. Our margin on these videos doubled (or better), from low or even negative percentages.
My role
I owned the business aspects of applied ML. The question is always “will the margin benefits be great enough to offset the engineering cost, and will we get there quickly enough for it to matter”. I performed that analysis and managed ongoing cost/benefit accountability across the tech org.
3D point cloud labeling tool
Build a tool for the hardest kind of annotation
LiDAR requires the thorniest and most time-consuming kind of annotation. Localizing objects in three dimensions requires highly specialized user interfaces, and they also have to deal with massive amounts of data, even for a single frame. We developed an optimized tool from scratch with a wide range of features purpose-built for efficient and accurate annotation at scale.
Key outcome
This opened a whole new market segment for us in autonomous vehicles etc. With this tool, we could improve margins without sacrificing quality, and win more business by pricing aggressively without losing money.
My role
I oversaw product management and design, ensuring delivery of a capable product that met business goals, while leaning on SMEs for deep expertise. I also launched the product, including sales and customer success kickoff and training.
Product career ladder
Give employees a path for advancement
In an early-stage company, it's nice to avoid too much formality, but as you scale, the product team needs a path for growth. The company had never had one.
Key outcome
The advancement tracks I laid out and negotiated with HR were well-received, and contributed to an overall 78-point improvement in employee NPS.
My role
I defined and delivered the career ladder, and worked with human resources to ensure it aligned with existing designations and salary tiers.
A true MVP
Clinch a big renewal while validating new product ideas
Marty Cagan says beware of “specials”, but what if one of your biggest customers was willing to pay for one and you could also use it as a low-risk/low-cost way to test the concept of a new product for an adjacent market?
Key outcome
We built a very focused app in 8 weeks, pleased Walmart, renewed them, made their users more effective, and learned a great deal about how a “data explorer” might fare in the market.
My role
I conceived of the solution, worked with the enterprise account manager to validate it with Walmart, sourced a nearshore development firm, defined the API and user interfaces, and saw the project through to completion by training users.
ML for accuracy and speed
Make an ML model that outperforms humans on color labeling
We needed to label garment colors in fashion catalog images, on a short timeframe with tight margins. We developed a model that outperformed humans at telling light gray from silver (and other fine distinctions).
Key outcome
We completed the labeling with far greater accuracy and speed than with people alone. Faster = better margins, and applying ML was the only way we could make a profit on this contract.
My role
I made the key go-for-it decision, and acted as scrum product owner to get it into production.
Price modeling
Determine how to price a new product
As we introduced a new self-serve version of our platform, we had a problem: billing for data annotation is hard because it's tricky to define what an annotation even is. We side-stepped the thorny problem by billing like a utility.
Key outcome
The simple billing scheme in the self-serve platform reduced friction in the buyer journey and reduced support headaches, leading to an increased likelihood of both purchase and renewal.
My role
I performed value metric analysis to determine feasibility and buyer appeal. I then advocated for the best method with executives and peers.
Enterprise reference app
Make API customers successful in days, not weeks
Humanify was a rich and sophisticated platform, but it had no “on-ramp”. We built a rocketship instead: Clarity, a mobile reference implementation to accelerate successful integrations by demonstrating best practices and API calls.
Next-level web analytics
Give users fixes, not just metrics
When your e-commerce site slows down, you lose money by the minute. You need a fix, not forensics. Instead of just a web analytics dashboard, we built one that makes suggestions in addition to presenting charts, and it also leveraged unique WordPress-only metrics.
Key outcome
In addition to differentiating the WPEngine experience from competitors’, this became marketing's best-performing lead-gen tool.
My role
I worked with the founder to clarify the initial vision, researched APIs, managed integrations with our partner NewRelic, defined the product's functionality and user interfaces, and drove stem-to-stern launch activities.
Data-informed decision making
Analyze and visualize usage metrics for better product decisions
Spiceworks was introducing a new product for their enterprise community users, and needed to make some foundational decisions.
Key outcome
We were able to tailor the new functionality to optimize for predicted demand.
My role
I crunched numbers and created novel visualizations, then used them with executives and other product managers to understand and advocate for the real voice of customer.
A true market pivot
Thrive in a downturn by changing product focus
Pluck established itself early as a leader in the new social marketing space, with a platform for integrating social features into high-traffic sites. But as the economy changed for publishers, we needed to find new buyers.
Key outcome
On a multiyear journey, we grew retail and brand revenue from zero to over $16 million; it became the core of our business.
My role
I inherited Pluck from the great product managers before me, but as director, I had responsibility for everything (the capabilities, UX, APIs, quality, integrations, technical documentation, and more).
2004-2007
Senior Web Engineer, InCircuit Development Corp.
Programmed interface and application layer for a SaaS-based asset management system used by large institutions such as the Louisiana state government and University of Texas. Tech included Java, HTML, CSS, and JavaScript.
- Completely revamped an application layer written in ColdFusion without a DRY approach; saved thousands of lines of code and untold time lost to duplicate maintenance.
- Helped in a total agile transition (implemented Scrum).
- Hired, trained, and managed a second web developer.
1998-2004
Full-stack web engineer (contract)
Houston. Contract developer with jobs lasting from 3 weeks (Time Warner Cable) to over two years (J. Ray McDermott). Worked through TEK Systems much of this time. Other clients included Dow (Union Carbide), AIM Funds, Schulmberger, Boeing, Webvertising. Tech used included SQL Server, ASP (classic!), ColdFusion, Visual Basic, HTML, JavaScript, CSS.
- Lead developer on a bespoke system used for J. Ray McDermott’s multimillion-dollar bids. Reduced the cost of developing and tracking bids by 5–10%.
- Modernized and added interactivity to J. Ray McDermott’s corporate web site.
- Refactored a greater-than-100,000-page training website for Dow. Reduced surface area for QA and maintenance by one third, lowering TCO.
1995-1998
Web developer (freelance)
Houston area. Part-time designer and developer of “brochureware” websites at a time when that meant lots of <TABLE>, some GIFs, and a little JavaScript.
1995-2000
Music teacher
Private trombone teacher for several school districts south of Houston, including Lake Jackson ISD.
- Within a year of starting, my studio was performing as well as the powerhouse Pearland ISD system.
- Wrote my own method book.
- Had sixth graders familiar with fifth position and F♯. Fellow trombonists will understand this is kind of a big deal.
Dec. 1994
B.S.—Music
Texas A&M–Commerce (trombone performance)