Software feature development for drones

Technologies involved: Go lang, Docker, REDIS, Kubernetes

I helped R&D Team at Nokia in the development of a scalable task manager which balances the loads for the drone software and flight controller with task scheduling features

Development of Robot fleet management and Solutions for enterprises

Technologies involved: Java, go Lang Docker, Memcache, Cassandra DB, REST, gRPC, TELNET

I helped R&D Team at Nokia to build and expose stable APIs for other microservices with respect to autonomous robots.Also, I was invloved actively in interfacing with 3rd party robot vendors and in retreival of telemetry and miscellaneous data from robots

The initial phase of this project was the development of autonomous indoor transportation solution for enterprises where a user could register with his face, location and request an item, then the robot was capable of moving from its station, picking the object and delivering to the user by identifying the user with facial recognition

The second phase involved autonomous trash management where a robot autonomously unloads trash from bins if a threshold is reached and dumps at a garbage collection center. I was mainly responsible for handling telemetry data, and the robot's miscellaneous data such as speed, direction etc and also developed a mediation interface for communication with various 3rd party vendors as we managed the fleet of robots with varied interfaces

Object detection and tracking using convolutional neural networks

Technologies involved: Python3, JUPYTER notebook, CUDA, TensorFlow, Keras, CUDNN, RESNET-100/150, RETINANET, CSV and NVIDIA V100, 180 Ti and Jetson TX2 hardware

I helped R&D Team at Nokia in the development of Convoluted Neural Network for object tracking and detection of Human beings and vehicles from very large heights using TensorFlow and Keras. I was also involved in data and model preparation along with training on GPUs. I have also worked with JetsonTX2 for implementing analytics and computations on the drone

Intra-Nokia Hackathon finalist

Technologies involved: Azure cloud, Java, Mongo DB, REST, gRPC

Traffic congestion is a common hindrance which is caused mainly due to automobiles parked in “No Parking” areas knowingly or unknowingly. Analogously, the police vehicles need to monitor these areas which further adds pollution and congestion into the system. Therefore, we proposed a Deep Learning based solution which can monitor vehicular movements. This involved vehicle and number plate recognition through which the owner’s details were fetched, and warning messages were sent after a certain time specified by the governing body. Meanwhile, the local police authorities were notified about the wrongly parked vehicle, allowing them to easily tow it, thus saving a lot of human and fuel resources which also in the process reduces the pollution. A Graphical User Interface was also provided at the local police stations to display various traffic violations and status across the city.

Runner up in an innovation campaign 'i-invent'

Technologies involved: Python, Angular 7, CSS, Bootstrap, Google cloud, Docker, Cassandra DB, REST, gRPC and Raspberry pi Hardware with PIR sensor

I led a small team where we developed a platform for resource management in refrigerators and food containers which detects the food ingredients and the food quality using Artificial Intelligence. A GUI and Android application was made available which allowed a user to view the status of the items with nutritional information and also suggested recipes with the available ingredients. The platform also notified the registered users if the items were at the brink of getting spoilt.

Stationed Automotive’s Surveillance of Streets for Smart City

One of the main reasons for traffic congestion is due to automobiles being parked in no parking zones across the city. The local police authorities go on patrols to inspect and tow the vehicles parked in no parking areas. This is a tedious process for them and results in the wastage of fuel resources. Since many areas of the city already have CCTV cameras installed, these cameras can be used to extract information about the parked vehicles. These cameras can be used to locate the vehicles in the no parking zones. Using Deep learning-based Object detection models, the automobiles can be detected, and the number plate can be read using an ALPR system. The number plate details can then be used to fetch the details of the owner from the RTO Database. The owner can now be warned by sending messages to his/her contact number via any messaging platform tied to a carrier service provider. On the other hand, the local police station can be equipped with a GUI display with the details of vehicles identified in no parking area. The system first warns the user to move the vehicle and if the vehicle is still stationed at that spot, a penalty is levied, and a ticket is registered at the nearest police station identified using the location of the CCTV camera and the owner is intimated about the same. If the vehicle is not moved even after the second warning, towing procedure is initiated at the local police station near to that location. Thus, this system reduces the burden on traffic authorities and helps in reducing the traffic congestion along with pollutions caused by the automobiles.

Voice Over WiFi Based Smart Wireless Notice Board

The notice boards are being handled manually in many applications. In order to put up notices on the notice board lot of resources such as paper, printer ink are wasted and addition it consumes a lot of time to make notification. In this project we have proposed a system to wirelessly transmit short notices using campus Wi-Fi to reach students quickly in the classrooms. The proposed system is low cost and energy efficient system as employs Raspberry pi controller to receive notices and display on LCD or terminal.

Channel quality prediction using Machine Learning techniques

Technologies involved: Java, Python, Android Studio, Google cloud

The project aims at developing a machine learning based technique for prediction channel quality measurements. Modern User Equipments (UE) often have multiple wireless interfaces such as 2G/3G/4G/5G cellular interfaces or WiFi. In the context of newly emerging intelligent transportation systems and the increasing demand for Quality of Service (QoS) for modern applications, the ability of a device to predict channel fading becomes essential. With the proper prediction technique the device would be able to switch to a more favourable connection before deep fading occurs and thus preserve the overall quality of experience

Smart Car Body (SCB) AI Platform - Volkswagen AG

Technologies involved: AWS (Kinesis, Lambda, SQS, S3, Athena), Python, Docker, Kubernetes, CloudFormation, AWS CDK, CI/CD

As Product Owner and Solutions Architect, I led the technical execution of the Smart Car Body AI Platform for Volkswagen's production data science operations. The platform processes 10GB+ of daily streaming data from manufacturing facilities to improve dimensional precision in automobile production.

Key achievements include: architecting event-driven microservices on AWS infrastructure, designing real-time data pipelines with Kinesis and Lambda, implementing the patented Coupled Process Analysis (CPA) algorithm for error detection, and delivering EUR 1.2M annual business value by streamlining manufacturing processes and enhancing quality control. The platform operates on VWS2.0 AWS infrastructure with automated CI/CD deployment.

Workshop on Raspberry Pi and IoT

Technical talk on Wireless Sensor Networks and IoT

Technical talk on Robotics and IoT for Faculty Development Programme at FISAT

Technical talk on Cellular IoT

Technical talk on IoT and it's Applications

Technical talk on IoT and Introduction to Deep Learning at Alva's Institute of Engineering and Technology

For projects, ideas, collaborations or anything else don’t hesitate to get in touch. Or just drop by and say hello. Coffee / Beer is on me :)