Retrievr: A smart Alexa Skill to locate lost items

Amazon Alexa + xgboost + BLE to find lost items around the house

MM

Built in collaboration with Terry Rodriguez

On many blogs and articles, we've found that many wished Alexa could help them find lost items around the house. We took on the challenge and hacked a BLE beacon to help us track a set of keys. We also built a model to help us estimate in what area of our apartment the keys might be.

Cellular Connected Autonomous AR Drone 2.0

Hacking an AR Drone 2.0 to fly autonomously using a Raspberry Pi Zero, and Hologram Nova.

MM

This project won Best in Remote Controlling in Exploring Cellular IoT with the Hologram Nova and RaspberryPi contest!

Built in collaboration with Terry Rodriguez

On this weekend project, we hacked an AR Drone 2.0 to navigate autonomously and added a cellular kill switch to override the drone's programmed navigation.

PoochPak: A smart dog wearable

A smart dog wearable for object recognition using YOLO, reading biometric data, and sending SMS notifications.

MM

This project is featured in Hackster.io's hand picked projects of the week and won Best in Asset Tracking in Exploring Cellular IoT with the Hologram Nova and RaspberryPi contest!

Built in collaboration with Terry Rodriguez

Diving into the space of smarter wearables, we wanted to take advantage of the dog experience to explore new ways of deploying cellular technology for remote object recognition and pet tracking.

Build A Raspberry Pi Photobooth

A fun diy gift using raspberry pi and peripherals

MM

Built in collaboration with Terry Rodriguez

For this holiday season we decided to make our loved ones a homemade gift. My parents are big fans of taking selfies, so we thought a digital photoframe would be something they could enjoy. We stepped it up a notch and added a selfie camera so their selfies could be incorporated into the photo slideshow automatically. Here is the walkthrough of how we put this thing together.

ML Tools: A little library for faster experimentation

Making experimentation a lot faster and easier to keep track of

MM

Don't be this dog when running your machine learning experiments! As a beginner it can be hard to keep track of all the things you're doing. With the noob in mind, I created a small python library to condense a lot of the code I was using into a few classes and functions that can be used and modified over and over. This is useful for the more seasoned data scientist as well, making experimentation a lot faster.

Neural Module Networks: A push for general AI

Some thoughts on the new BAIR lab blog and experimentation

MM

The BAIR lab has recently gone live with their new blog and their first post was on Neural Module Networks, a new kind of neural network designed to tackle more complex problems by chaining pieces of different neural nets. O'reilly showcased in one of their newsletters Google's "One Model to Learn Them All" where they train a large neural network on many different types of data, combining different neural network architectures to create a model to solve problems across multiple domains. These two groups are trying to reach for the same goal of developing more flexible models that can answer complex questions fast, without having to train multiple networks for very specific cases.

Autoencoders for dimensionality reduction

Autoencoders for dimensionality reduction and summarization

MM

Continuing to explore what tensorflow has to offer, I've been playing with autoencoders with the Kaggle Mercedes-benz dataset (yet again). Having a small dataset with a large feature space, it can be beneficial to reduce noise and potential to overfit by condensing the most important information and patterns into a smaller feature space.

Deep Learning with Tensorflow

Experimenting with a basic Multilayer Perceptron model on a small Kaggle dataset

MM

Thanks to the release of TensorFlow in late 2015, many of us who are not at the cutting edge of deep learning have been able to dig in and begin learning about the subject. This is a short post on a tutorial I tweaked to train a basic multilayer perceptron model on a small Kaggle dataset.

Project Series: Tinder for your Craigslist apartment hunt

Part I: A twist on a Slack bot craigslist tutorial.

MM

It is spring time and the apartment search frenzy is about to start in the Bay Area. Trying to find the right place to call home can be a nightmare in such a competitive market. In this project we will use custom filters and eventually incorporate a recommender system trained up on our taste to find the right spot. This project also takes inspiration from the dating app world with a tinder inspired interface to find that match.

First Post

A brief introduction.

MM

Welcome! This blog will serve to document and share insights and interesting projects during my journey deep diving into AI.