Over the past few years, I have been creating prototypes of ideas that I find interesting. Mostly to see what I can build, and to understand a new technology. Sharing some of the prototypes that I had worked on:

​

• AirBnb price optimization using neural network -- {Deep Neural network}

• Mellowain : Auto Checkout Without Standing in the Queue -- {OpenVX, Deep Neural Network, OCR, Nvidia Jetson TX2, Load Cell}

• 3D Image Sampler -- {OpenCV, Relays, Arduino, Load Cell, Barcode Reader}

• Nutridiet -- {Random Forest classifier}

• Business Card Reader -- {OpenCV, Image Processing, OCR}

​

​

AirBnb Price Optimizer 

(Deep Neural Network)

​

This is available on colab. As a host you always want to offer great home to guests and in return expect best price for your house. But to find best price in your vicinity and for your type of home, is not always easy, so you strike a guess price and put it up. But that price can be either less for the services you offer or it can be more and hence the house occupancy is not full all the times. This puts you in never ending 'hunting' mode for the best price that will never be a 'sweet' spot because it is constantly running target. 

 

This is when you need a automated system that can find a optimum price for your home considering all the amenities and services you provide to guests. This can also give you very good intuition as to what you need to fix (amenities and services) in the house to earn more.    

​

​

Mellowain : Auto Checkout Without Standing in the Queue

(OpenVX, Deep Neural Network, OCR, Nvidia Jetson TX2, Load Cell)

​

The source-code of the project is available on request. This idea we pursued and reached to the Techstars final list, but got rejected in the final round. 

​

"We wanted shopper to shop & not stand in queue, Grocery stores to offer realtime services to shoppers. 
We believe shopping is an experience, and to make it fun and delightful for everyone."

​

The demo video can be found here. 

​

​

​

 

 

 

 

 

 

 

 

 

 

​

 

3D Image Sampler

(OpenCV, Relays, Arduino, Load Cell, Barcode Reader)

​

If data is oil, machine learning is oil refinery!

 

To generate your own image data for a specific object/product/item. We have created this project, that captures 360 degree images of an object. The work is based on the image capturing details given in the paper

This sampler is meant for image sampling that can be later used for training an image classifier. To get best quality and appropriate image samples the paper concludes following:

• Azimuth - Object should be revolved 360 degree at least 24 times in equal angle. This means 360/24 = 15 degree.

• Elevation - In a circular fashion object should be captured in four verticals i.e. 90 degree / 4 = 22.5 degree.

​

Code repository can be found here and demo video. 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

 

Nutridiet

(Random Forest classifier)

​

USDA Dataset covers most of the fruits and vegetables nutrition composition. It also gives daily Dietary Allowance and Adequate Intake for a day of all age groups and gender. I thought of making a automatic caretaker who can keeps track of your Nutritions and notify you when you consume more or lack of any particular nutrition.

 

Before training the dataset was cleaned a bit and trained on Random Forest (RF), the dataset was thankfully in proper CSV format. I trained it on my data with following features: 

​

​

​

​

 

Beauty of Random forest is that it hardly overfits and works even on single row on information about a food item.  

​

​

Business Card Reader

(OpenCV, Image Processing, OCR)

​

Assume, I am a business person who meet many people from my industry and as a ritual we all share our business cards but hardly enter their details on our phone book and when need arises, we hunt for his/her business card. Would it be better to have a model that extracts all the text on the business card and feed them to my phone book. 

​

Next step is to have adaptive functions for thresholding so that no incorrect data is entered into phone book. The algorithm also corrects the orientation of the card by rotating it. OCR is very sensitive to alignment of the image. 

​

The code is on my colab here. 

​

​