Cambrio builds AI (automated intelligence) tools that accelerate your capability to prototype, benchmark, and operationalize analysis workflows. We specialize in complex domains like: life sciences, geospatial, scientific sensors, and healthcare, where deep expertise is required to understand context and "see" things or "read" signals. Because the demand for attention is far greater than the supply of available experts, important patterns are routinely missed. From computer vision to language to multi-modal scientific data, Cambrio's automated workflows are the best way to score, prioritize, and scale the impact of your analysts and subject matter experts. If you are looking for ways to automate your own analysis workflows and build data leverage, reach out! hello@cambr.io We love to connect people who are attacking hard, impactful problems.

Developed new technology to keep indico on the bleeding edge of AI/machine learning and deep learning (convolutional neural networks, recurrent neural networks, deep vs. wide models, attention, fine tuning, transfer learning, etc). Specifically, I focused on new ways to represent and learn from data when you don't have millions of labeled examples. To share knowledge/expertise and build community, I spoke at events and regularly wrote articles on topics ranging from the fundamentals of machine learning to practical tutorials using Tensorflow.

At a time when wearable devices were all about quantity/counting stuff (steps, miles, heartbeats), Aidetic developed algorithms to measure the shapes of human movements in real-time, using the sensors that were already installed in 1B+ mobile devices. It was a pretty classic early-stage experience: built hardware + software prototypes, designed machine learning algorithms to work on streams of sensor data, validated predictive models against real-world movement behaviors/data, grew relationships with users and investors, developed strategy, and tried to keep the wheels on. Shut it down to focus on deep neural networks.

I led data science projects as part of the Strategy, Innovation & Analytics organization. Operating as internal consultants, we partnered with business unit managers to understand how business operations could be improved, using data + predictive models. Example: using weather data and historical claims volumes to predict where appraisals will be needed following major storms.

Consulting tech expert for science, engineering, behavioral economics, and computational stuff; scope spanned pre-sales to business development (i.e., partnerships, testing new kinds of products) to actually implementing version 1.0 for new software features. In helping organizations solve technical problems using open innovation, I had a hand in many of the world's toughest technical challenges, discovering a great diversity of innovative solutions. Projects were covered by Nature, New York Times, The Economist, NPR, Fast Company...and even highlighted (positively by both sides!) as an example of thrifty innovation during the Obama/Romney presidential debates. Example article (via Fast Company): http://tinyurl.com/7ne7fmm. I also led enterprise-grade workshops to coach various organizations how to formulate their problems to be solved by smart people on the internet. Audiences spanned 1-100+ people, from research scientists to C-level execs. When clients wanted more, I developed the world's first certification program for open innovation (i.e., "train the trainer"​) and rolled it out to U.S. government, Fortune 100 companies, and top consulting firms.As the lead data scientist, I also owned the company's agenda for data-related and analytics opportunities, working with the CEO to validate company strategies for data science, big data​, and machine learning as quickly and frugally as possible. The most visible result was the world's first commercially successful platform for online data science competitions---over $1.1M awarded to solvers around the world to date. It worked surprisingly well, and inspired a niche market (Kaggle, etc) that thrives today.

Before "data science" was a thing, Aentropia was my vehicle to explore challenging computational problems, provide data/consulting engagements, and develop intellectual property.

Using huge amounts of proprietary and public data (hundreds of genomes, all kinds of "omics" data), we explained how T-cells behave during various autoimmune diseases. Our model showed how key molecular networks are competitively balanced in the healthy state, and how disease states happen when inflammation pathways disturb the healthy balance.I was also really lucky to be pulled into a unique collaboration, where we shared 100% openly with academic collaborators, at a time when big pharma was notoriously proprietary! Our successes, in terms of both relationships and scientific results, became a prototype for Pfizer's industry-academic drug discovery partnerships, consumated by $107.5M investments with our collaborators at Washington University and UCSF. Wall Street Journal coverage: http://tinyurl.com/6nhh2dx

Dissertation gist: We discovered a solution to a long-standing paradox of molecular biophysics, by using billions of experimental observations and statistical physics to show how protein molecules are predictably flexible. In hindsight, this strategy of using distributed computing tricks to apply a mathematical model to many many observations was an early example of using "big data" to solve a scientific problem. But nobody called it that. They said "Dude, are you using the whole cluster, again?"Other threads: (1) Validated the accuracy of constrained search algorithms for molecular design, by showing how predictions from decades before were at least as accurate as recently published atomic structures (sub-Angstrom accuracy!). (2) Many hours in "wet lab" isolating molecular targets, cloning genes, purifying proteins, synthesizing peptides, babysitting cell cultures, testing combinatorial molecular libraries using high-throughout robotics, troubleshooting mass spectrometry machines, accidentally scaring the late-night hospital janitorial staff ("is he dead or sleeping?"), developing assays, etc. (3) Explored systems biology and the modular two-component systems that bacteria use to adapt to environmental conditions. (4) Built high-performance computer clusters from cheap consumer parts. (5) Designed a rigidified molecular template to mimic peptide side chains in a way that could exploit combinatorial chemistry, so we could test entropy/enthalpy compensation with thousands of experiments at once. (6) Taught classes (biomedical engineering senior design, molecular design labs, computational chemistry, computational biology). (7) Mentored and learned from some really bright classmates and colleagues!