Alfredo Andere 🦖

❖ Created a webpage that abstracts computer vision and makes automated image classification accessible to anyone. (flash-cv.com)
❖ I built the back-end of the project using PyTorch. I leveraged transfer learning on SqueezeNet (Landola et al. 2016) by re-training it for any other image classification task.
❖ Won Cal Hacks, the world’s largest collegiate Hackathon.

Using Pandas and NumPy I manipulated stock data from the S&P 500 to create half a million features and labels. The features were data on the last thirty days of a stock. The labels were one-hot encoded vectors of the stocks price going up or down in the next 24 hours.

Using Tensorflow I built a Deep Neural Network and trained it using the features and labels. It predicts if the price of a stock will increase or decrease in the next 24 hours. I achieved a 2.22% edge on the market. I… Show more

Using Pandas and NumPy I manipulated stock data from the S&P 500 to create half a million features and labels. The features were data on the last thirty days of a stock. The labels were one-hot encoded vectors of the stocks price going up or down in the next 24 hours.

Using Tensorflow I built a Deep Neural Network and trained it using the features and labels. It predicts if the price of a stock will increase or decrease in the next 24 hours. I achieved a 2.22% edge on the market. I continuously seek to improve my work.

(edit: after learning more ML I learned that this result was “too good to be true” and actually my test / train sets were contaminated because I split them randomly instead of chronologically) Show less

Built bottom-up a voice-controlled car using the Arduino IDE and a Texas Instruments Launchpad.

Assembled a mic-board, built the whole circuit system, created a closed-loop feedback controlled algorithm, and used least squares to optimize the performance of the motors.

Currently working in voice-detection and word-classification for the voice control feature.

Participated in a 28 hour hackathon organized by General Motors.

My team built and trained a Convolutional Neural Network which takes in a customer complaint and classifies it under one of 200+ categories.

We used 100,000+ samples of company data. Processed sentences using Pandas and NumPy and used Word2Vec to capture context. Used TensorFlow to build and train the Convolutional Neural Network (CNN).

Developed the back-end of a mapping application of Berkeley using Java, Apache Maven, and images downloaded from OpenStreetMap.

Created a quad tree to filter through thousands of images and raster the appropriate ones into the front end. It can handle features such as moving the screen by dragging, zooming in/out, and double clicking two points to find the shortest path.

Employed A* (A star) algorithm using a priority queue to find the shortest path between two given points.