From the past one year, I have been doing research with Prof Min Xu at School of Computer Science, Carnegie Mellon University in Video Anomaly Detection, specifically using Semi-supervised Learning, Knowledge Distillation and Feature fusion. Our method beats the current benchmarks on UCF-Crime and ShanghaiTech using intuitive and scalable techniques. Our work is submitted to the CVPR'24 conference. 🤞
I interned at Rubrik last summer as a Software Engineer Intern, where I implemented Multi-Node Scaling in RSC-P Cluster (On-Prem version of the Saas product, used mainly by govts. and major banks). In the process, I got an amazing experience with Distributed Systems, Golang and Kubernetes. I also developed a CLI for the clients to add/remove new nodes very easily. Received a full-time offer from Rubrik. 🥳
I am doing my undergraduate thesis under the joint supervision of Prof John Jose and Prof Sukumar Nandi in the Detection, Analysis and Mitigation of Blackhole Trojan attacks in Network-On-Chips. We are simulating the attack in gem5 simulator using C++ and Python. Our impact analysis currently shows the effect of such attacks at the processors, cache and network levels, and we will be working on developing an efficient mitigation scheme in the upcoming semester.
Previously, I did a Research Internship with Dr Rohitash
Chandra and Dr SS Vasan under the joint collaboration of
UNSW-CSIRO Australia, particularly in the analysis of variants of SARS Cov-2,
with special focus on coming up with predictive results on mutation tendency of
aminoacids.
I look forward to interacting with people and participating in team activites.
In my leisure time, I pursue sports, video games, quizzing, and also write blogs on my personal blog (previously on Quora & Medium).
I am an avid YouTube consumer as well :p
Please feel free to check out my resume
and drop me an email
if you want to chat with me!
Volunteered as Subject Matter Expert in Physics at Edvizo, teaching high-school
students.
Nov '20  
Started studying B. Tech in Computer Science and Engineering at IIT Guwahati!
Bachelor of Technology in
Computer Science & Engineering Indian Institute of Technology Guwahati
Nov '20 - May '24 (expected) CGPA: 9.07/10.00 Minor in Robotics and Artificial Intelligence
Positions of Responsiblity:
Contingent Leader of IIT Guwahati, Inter-IIT Technical Meet 12.0, IIT Madras
Treasurer and Ex-Technical Co-Ordinator, Computer Science & Engineering Association
(CSEA)
Research Internship Carnegie Mellon University, USA May '22 - Present
Advisor: Dr. Min Xu
Submitted our research paper to the CVPR 2024 conference, with an improvement of 1.36% over the state-of-the-art on the Weakly-Supervised Video Anomaly Detection task on the UCF-Crime and ShanghaiTech datasets.
Proposed a novel Temporal Aggregation Module which is based on Disentangled Cross Attention to aggregate the representations of multiple backbones and distil the knowledge using Bi-level Knowledge Distillation.
Software Engineering Internship Rubrik Inc., India May '23 - Jul '23
Team: Legion (Polaris Platform)
Implemented Multi-Node Scaling Support in RSC-P Cluster(Rubrik Security Cloud- Private) using Kubernetes.
Developed a Command Line Interface (CLI) in Golang for the users to add/remove nodes to an existing RSC-P Cluster, view cluster status, and perform live updates using Gravity.
Devised and implemented a Leader-Worker architecture to pin the stateful services on the leader node, while scaling the stateless services automatically to the newly joined nodes.
Undergraduate Thesis Project IIT Guwahati, India Jul '23 - Present
Advisors: Dr. John Jose and Dr. Sukumar Nandi
Proposed a new Black Hole Router Attack threat model on Network-on-chip on the cache, processor and network levels.
Analyzing the impact on L1 and L2 cache misses and penalties, CPU-wise IPC values and average packet latency using the gem5 simulator, and coming up with possible detection and mitigation techniques for such attacks.
Research Project IIT Guwahati, India Nov '23 - Present
Advisors: Dr. Arijit Sur
Working on Adversarial Defense on 3D Human Pose Estimation models using Diffusion Models to implement a generator-discriminator-classifier training paradigm to train a robust downstream classifier against adversarial perturbations.
Research Internship University of New South Wales Sydney & CSIRO Australia Nov '21 - May '22
The project is a virtual tour of the Majuli Island in the Brahmaputra River, which is known for its rich culture and biodiversity. Through the use of 360 degree images and information in the form of annotated locations and video playback, the product aims to give a holistic tour of the key parts to the user.
We went through the various necessary steps in software engineering life cycle, including analysing user requirements (based on which the app was divided into 3 separate tours), writing an SRS Document, building the prototype, and making the final app. With around 300 images and 14 videos used, the final app came out to be around 1.3GB.
We built a complete NanoC Compiler(subset of the C language) from scratch using Flex, Bison and C++. We first built a two-pass assembler, its syntax tree and then finally the compiler to bring it all together. It supports basic data types, arithmetic and logical operations, conditional statements, loops, functions, arrays, pointers, structures, and recursion.
Hybrid of Shortest-Job-First and Round-Robin based Hybrid Scheduling Algorithm, and tested it for CPU and I/O bound tasks in xv6 - a UNIX based OS. It sped up the CPU bound tasks by pre-empting at the end of time quanta and rescheduling the best-fit process.
Memory is allocated only when the process actually needs to use it instead of declaration.
Provisioned creation of kernel processes and dynamic page swapping for processes using a circular queue (LRU policy).
Analysis of the Deduplication feature of the ZFS file system and the Large File Creation feature of EXT4 file system. We ran multiple workloads on instances of these file systems. We compared the space efficiency and creation times for the two file systems.
We built a Smart cycle with features like an Anti-theft alarm, Find Your Cycle (from a mobile app) and Automatic front and taillights.
We used a ESP8266 NodeMCU with MPU6050 Accelerometer and Gyroscope, GPS Neo 6M I2C, Grove v1.1 Light Sensor, Piezo Buzzer and LED strips.
Additionally, we built a mobile app using Flutter to that notifies whenever someone tries to move a locked cycle using MQTT Protocol, its current location and allows remote locking/unlocking of cycle.
A Banking System using Client-Server TCP socket programming, implemented from scratch in C, which has file-based login and stores customer account transactions.
Implemented a reliable sliding window transport layer on the top of the UDP Protocol, having features to handle packet drops, packet corruption, packet reordering, error-detection in packets and multiple outstanding packets.
Simulates the Bash shell using execv calls in C with the provision of redirections, single-level piping, pipe-separated commands, environment variables, cd command, history.
Route Planning and Optimization with Volume Estimation for Last-Mile Deliveries
Performed volumetric weight estimation of objects using a stereoscopic RGB Camera setup and developed an end-to-end physical system to measure the dimensions of the object with very low latency and 95% accuracy
Used Deep Neural Networks to classify the shape of the object, for background removal to estimate Pixel Per Metric ratio, and using Similarity Scaling and Background Averaging to predict accurate object height
Optimized the Last-Mile-Delivery problem by devising Optimal Rider Routes, incorporating Dynamic Pickups and forming rider-bags using 3-D Bin Packing heuristics. Improved scalability by using a iterative Sweep Clustering Algorithm.
Bosch's Model Extraction on Video Classification Models
Deployed Gensim Word2Vec model with custom parameter tuning, interactive semantic similarity plots, and training-on-the-fly features. Option to upload custom text corpus or tabular conversational data also provided to the user.
Performed Dimensionality Reduction on the Word Vectors to map them to 2D and 3D space using PCA & TSNE.