Zeitle

Azure Cloud Architecture
Chatbot Image Analysis on Cloud using AI
Chatbot Interaction
Image Segmentation Results
Deep Learning Model Accuracy

Inspiration

Tropical countries suffer from sudden outbreak of diseases and having limited number of medical facilities resulted in congested and poor medical services and dengue is one of the most common infection disease among them. According to WHO, 390 million people suffer from dengue and only 90 million people have access to right medical facilities. Having team members from tropical countries already experienced the challenges of poor medical facilities, we took steps to innovate this medical experience.

What it does

Our solution enables the medical practitioners and patients to perform self-service virtual assistance regarding the cure of the dengue disease with the application of artificial intelligence, natural language processing, and computer vision. Having self-service capabilities, our innovative solution enable the patients to chat with the virtual assistant and get the medical scanned by the artificial intelligence models for several diseases detection.

How we built it

This digital virtual assistant is realised on Microsoft Azure using various cognitive and AI services including Azure Bot Framework, Azure cognitive vision/speech services, Azure Blob Storage, Azure LUIS, Azure QNA, and deep learning models using pyTorch and openCV. The deep learning models are deployed on the Azure cloud with virtual machines having the GPU compute power to reduce the time of analysis. We used Apeer framework from Zeiss to assist data collection, annotation, and reproducible research for researchers and doctors for future research.

Challenges we ran into

We ran into several scalability challenges from training deep convolutional neural networks to running data pipeline on the Azure cloud. We faced compatibility challenges while integration various services including communication between bot framework, cognitive services, Apeer framework, and model deployments. Apart from that, we did some boring data labelling and transformation, along with scraping external datasets from outside sources.

Accomplishments that we're proud of

98% Accuracy with balanced Recall and Precision for Deep learning models for blood cells detection and diseases classification. Running end to end scalable solution on Azure cloud and the integration with various cognitive and deep learning models. Running models in the inference mode on the GPU compute virtual machines and integrating with chabot backend.

What we learned

Running artificial intelligence model in scalable manner along with hacking the model training time Integrating multiple micro services including cognitive services and deep learning models. Team work and working in intercultural team.

What's next for Zietle

Form a Start-up with the focus on helping tropical countries and structuring the Chaos! Start working on full-time as all of us are at the end of studies and looking for forming our startup.

Built With

ai
azure
chatbot
cognitiveservice
luis
python
pytorch
qna

Submitted to

ZEISShack2019
- Winner DIY robot, Philips Hue, Amazon Echo Dot

Created by

I worked on Histopathologic Cancer Detector challenge and developed a deep learning based solutions that employs Self Attention and Resnet modules to get validation accuracy of 0.99

Haris Jabbar
Developed Chatbot backend
Developed Cloud architecture
Integrated Deep Learning models
Integrated Cognitive services into backend.

Private user
My role during the hack was of a of a business person, designer and a pitch master.
Respectively I was developing the business model by performing the online research and torturing mentors with endless questions.
On the other side I was also responsible for creating the slides and delivering the pitch.

Andrey Bogomolov
I worked on:
1. Creating Deep Learning model for Cell Detection, and Nuclei Segmentation
2. Setting up pipeline for Apeer and publishing workflows.
3. Concocting business idea

...and having fun!

Sarthak Gupta
Data Scientist with expertise in Computer Vision and Natural Language Processing