Things I've made to put my dent into the universe.

I've worked on all sorts of projects over the years. Here are some of the main ones that I have been fortunate to lead or collaborate on. These span a variety of domains: outer space, hedge fund signals, automation of SaaS platforms, deep tech AI optimization frameworks, and RF signaling.

  • <Stealth Mode>

    Coming soon... :)

    (Private)

  • RTT

    A Contextual and Multi Armed Bandit Reinforcement Learning framework that is geared towards optimizing AI serving infrastructure in real time.

    (Private)

  • AMOS

    A neural network rebalancing framework that allows for optimal allocation of FLOPs within a model. Works based on curve fitting scale related parameters and then passing curves to a non convex optimization solver with Lagrangian multipliers.

    (Private)

  • Satellite Imaging Inference System

    An AI Inference system focused on the nuances of satellite imagery. This covers model compression, publishing, and serving. It is tailored / optimized for high resolution satellite imagery and is currently deployed serving super resolution use cases.

    (Private)

  • Attack Surface Management

    A platform that looks at cyber security from an internet observability perspective. We scan and monitor 4 billion ip/hostname/port combinations, 400 million domain names, full certificate spoof checking, and CVE monitoring. We then provide insights based on various seed ips/domains of your interest.

    acdsglobal.com

  • Deep Timeline

    In between jobs I was very interested in embeddings right as CLIP was coming out and vector databases were starting to become more mainstream. I built a prototype system that would continually index everything you saw on the web in vector space and then allow you to query for concepts, bringing up exactly where you saw similar ideas in a timeline view of your browsing history. I still love this idea, but did not have enery to pursue it further at the time as to do this well requires VC backing, which I was not in the process of doing at the time.

    (Private)

  • Eckleburg

    While at Ubiety, my flagship system was a streaming DAG based feature extractor that had a layer of multi task learning on top of it. This was to reverse engineer multi modal RF signals and classify unique electronics in real time. This was complete with a simulation environment, real time observability, and prod deployment.

    ubiety.io

  • DARPA LwLL

    In conjunction with DARPA and NASA JPL, I was tech lead of the core team for designing the overall framework for our AutoML system focused on semi-supervised and active learning.

    darpa.mil

  • Vision System for Mars 2020 Perseverance Rover

    At NASA Jet Propulsion Laboratory, I helped build the computer vision component of the autonomous navigation system for the 2020 Mars Perseverance Rover. This system was dedicated to estimating energy consumption of various subsystems that then fed into the larger control system. Some of this work ended up published here.

    ieeexplore.ieee.org

  • DARPA D3M

    In conjunction with DARPA and NASA JPL, I worked on the core team for the Data Driven Discovery of Models (D3M) program. This was geared to building an AutoML system that was composed of interchangable ML Primitives using metalearning feedback loops.

    datadrivendiscovery.org

  • Reinforcement Learning based Scheduling for the Deep Space Network

    At NASA Jet Propulsion Laboratory, I built a framework for general purpose scheduling problems (applied to NASA's Deep Space Network) with deep reinforcement learning as it's engine. This work eventually made it's way into a paper here.

    scholar.google.com

  • Telemanom

    At NASA Jet Propulsion Laboratory, I worked on general purpose anomaly detection system for spacecraft telemetry data. The methodology was eventually open sourced here.

    github.com

  • Radar Based Crop Yield Signals

    Back when the state of the art in Computer Vision was newly ResNet, we used radar imagery from satellite companies, profit shared, and sold signals to Chicago based hedge funds for region based crop yields around the world.

    (Private)

  • NLP Signals

    Back when neural networks didn't do text (pre transformers), I built a system based on the ideas of ULMFit to derive sentiment signals and create several indexes that we sold to various Chicago based hedge funds.

    (Private)

  • icd

    Tooling for interacting with ICD-9 and ICD-10 codes and comorbidity mappings. I get various doctors and researchers who reach out from time to time so I provide small feature updates for the greater good.

    github.com

  • fastteradata

    I had to interact with Teradata at a previous engagement and was frustrated with the lack of tooling and performance. So I built some tooling to make this process easier and it seems a handful of people forked and starred the repo so I guess it provided value for others as well.

    github.com

  • PadScouts

    Originally a suite of tools for real estate agents including, a CRM, a website builder, and a lead generation tool. Now a real estate brokerage.

    padscouts.com

  • RCMS Software

    An end to end platform to digitize and automate the back office management of construction companies. This automates invoicing, billing, purchase orders, equipment tracking, and fleet management.

    rcmssoftware.com

  • FirstClass Eventz

    A Transportation Logistics software platform that has operated events such as the Super Bowl and PGA Tour events for the past 6+ years.

    firstclasseventz.com

  • Augie Green Light

    An iOS app that allowed college students to organize private parties and events.

    (Private / Lost)

  • Catch! The Cat?

    My first app and introduction to programming. An iOS app where you catch falling cats from the sky and collect points.

    (Private / Lost)