I'm Anagha M Anil Kumar.

Summary.METAL focuses on optimsing search in multi-level indexed data structures, with non-affine accesses. Contributed to developing a new caching structure to accelerate non-affine accesses in multi-level indexed data structures. I have used C++ and LLVM IR to programme Gem5-SALAM as a hardware accelerator simulator for METAL. I have also used Gem5 to build customised cache modules to support METAL. METAL uses cache management policies called meta-patterns to perform application specific caching.
Language Used. C++, LLVM IR, Gem5, Python.

Summary. MarvinEye is a computer vision framework entailing 3 distance estimation algorithms (SDEC - novel, Triangle Similarity and Sensor Height), out of which one serves as an elector, which performs dynamic election and picks which of the other 2 algorithms is to be followed at that instant. Therefore, introducing reliability and robustness into the framework. MarvinEye gives the object it is integrated with a sense of position by estimating the distance to a pre-set visual aid, in a resource optimal and GPS independent manner. We have developed a new computer vision algorithm, taking inspiration from the DSMAC terminal way guidance algorithm for air artillery systems, adapting it to fit MarvinEye and it goes by the name SDEC- Spatial Distance Estimation and Correlation, where the distance moved is validated based on how much of the frame the object occupies- Clarity of Vision (CoV). The SDEC derives an equation for the best fit curve plotted using the CoV, through Polynomial Regression. The Sensor height and Triangle Similarity algorithms have also been used, after optimizing them to estimate distance. In order to increase the throughput, the code has also been accelerated on the GPU. The MarvinEye framework can be integrated into white canes for the blind, path following bots and in other constrained environment applications. It can also be integrated into Rovers to ensure smooth operation and movement validation. I have authored the paper and it has been accepted in the International Conference on Innovations in Information Technology 2020 to be held in the UAE. This project was done in association with the Centre for Advanced Systems - Parallel Systems Research Lab (CAPS-PSRL) in PES University under the guidance of Dr Rahul Nagpal.

Summary.Designed and implemented a MANET in a plug and play fashion. Tasks are distributed to clients in the cluster over TCP/IP, utilizing a star topology with 1 master node and several clients. is an application that can be installed on android mobile devices and also compatible on Linux systems, it functions over the LAN. Any user can assign work through the server, the tasks are partitioned (in order to exploit task-level parallelism ) and distributed over the client cluster based on the priority of the client using weighted round-robin load balancing, considering the device heterogeneity (utilizing client specs like number of the cores and clock speed currently available). A client can join the network at any point in time, the network handles both master and client faults and gets the network functioning again in a very short duration. If the master fails, priority-based election takes place (based on compute power) and if a client fails, its tasks are re-distributed to another client in the cluster (plug and play fashion).
Language Used. C++, C, CMake, Java.

Summary.Created a Database as a Service for the RideShare Application utilizing the Amazon Web Services EC2 instance. I created the RideShare Application, which was hosted on Amazon EC2 instance and an SQLite Database was used to store data. The application provided services to create a user, ride, add cab sharers to the ride, cancel a ride, delete a user. The user and rides functionalities were hosted as microservices on separate containers with a load balancer - NGINX. High availability and Fault-Tolerant Database as a Service was then created for this RideShare application. I implemented a custom database orchestrator engine that performed database read and write. RabbitMQ (Advanced Message Queueing Protocol) was used as the Message Broker. Round Robin Load Balancing was implemented in the master-slave configuration, using NGINX. Zookeeper was used as the cluster coordinating service, keeping a watch on all the nodes for faults. Automatic scaling was implemented, where if the number of requests increased, more workers were spawned, and workers were killed after a preset time if the requests decreased. Fault tolerance was implemented for workers and master if the master failed, a custom developed re-election was held based on priority and if a worker failed, another worker was spawned only if the failure did not take place due to scaling down. High availability was introduced into the application through zookeeper. The whole application was launched and managed on the instance in the form of docker containers.
Technology used. Python, RabbitMQ, Zookeeper, Docker Containers, AWS EC2, NGINX, Amazon Elastic Load Balancer.

Summary. Designed and Implemented a model using Arduino in Embedded C. The model made was made with consideration towards slow-moving traffic congestions and stray animals that may unexpectedly cross a road. The model is a 4-way traffic junction that has 4 signals placed individually on all sides. 2 infrared sensors were placed at a measured distance on one side of the road for each of the 4 roads. LED lights were set up to imitate traffic lights (red, yellow and green). The input for these LEDs was from the infrared sensors, on its assigned road. If the traffic density on a particular road (measured by infrared sensors) was high ( a lot of cars one behind the other), the light would switch to green and remain green till the traffic reduced. If all have equal density, the LEDs emulate a timed Traffic Signal in a pre-determined order. The Ultrasonic Sensors are placed at the Zebra Crossing and they can distinguish between a car and a person. This is done by using 2 ultrasonic sensors and also considering the current traffic light status. If the light is green and there is no vehicle in the vicinity, the ultrasonic does not trigger the sound alert system. The sound alert system is triggered if a pedestrian is crossing the road when a car is speeding at a certain distance from the pedestrian. The sound system acts as a warning to the driver, alerting him/her of the pedestrian. The sound system is also not triggered when the light is red and/or the cars have stopped to let the pedestrian pass. The LDRs were used in the project to automatically turn the street light LEDs on at nightfall.
Technology used. Arduino, Embedded C, Obstacle Sensors, Sound Actuators, LDRs

Summary. Designed the complete system starting from the ER Diagram, then coming up with the relational Tables. The Library Management System consists of 3 views- the Manager view, the Librarian View, the Member View. All 3 can view the list of books available across all branches and search through the same. The members can additionally reserve books, such they can access it as soon as it is available. The librarian can issue the books and return them for the members through this online portal. The Librarian can also add members and remove members based on the member's duration and renewal of membership. The manager can add and remove librarians in any branch. Databases were created in PostgreSQL for terminal execution of queries and MySQL for website execution of queries, to store all data which was hosted with a user-friendly front-end, designed using HTML, CSS, JavaScript. PHP served as the backend. The user can execute simple and complex queries through the website.
Technology used. MySQL, PostgreSQL, HTML, JavaScript, PHP, CSS, Apache Server.

Summary. Created a mini compiler for Java written in yacc and lex, to initially design the Context Free Grammar, which included constructs like multiple nested for loops, while,if-else, print statements, etc. Then generate the Abstract Syntax Tree from a Java class. Then the Intermediate Code was generated taking the AST as input to generate 3 address codes. The 3 address code was then taken as input for code optimization. The code optimization removed dead code, performed constant sub-expression elimination and constant folding. The code was then pre-processed before converting the optimised code to ARM Assembly. The pre-processing was automating the accumulation of all commands under the respective branches. From here, the Assembly code generated was ARM, this was achieved with python. This whole mini compiler execution was automated through bash scripts.
Technology used. Yacc, Lex, Shell Scripting, Python, Java.

Kepler Data Analysis - Data Science
India Tourism Website - Web Technologies, SQLite
Book My Slot - Android App, SQLite, Firebase
LBW Classification - Machine Learning
Best Before Notification - Android App, Firebase (Amazon - ACMS)
Book Recommendation System - Machine Learning, Web Technologies, SQLite
ODAM- Onboard Detection and Autonomous Maneuvering - Embedded, LiDAR processing (ICSS 2019)