Hi, I'm Sahil Dhull, about to graduate from IIT Kanpur with Computer Science (CSE) as my major. My interests span across Systems, Machine Learning and Robotics. Among sports, tennis, swimming and cycling are my favourite. And I'm becoming more of a fitness freak lately 😉.
If you want to know about my journey in detail, you can visit my blog.
Primary and Secondary Schooling Grade (Class X): 10/10
Abhinav Public School
Completed High School Grade (Class XII): 97.4% (CBSE)
FIITJEE Punjabi Bagh
Prepared for JEE Mains and Advanced AIR 27 in JEE Mains AIR 230 in JEE Advanced
B. Tech in Computer Science and Engineering CPI: 9.0/10.0
Obtained Pre-Placement Offer (PPO) from Adobe India Pvt. Ltd. after the internship
Awarded A* grade for outstanding performance in the courses Formal Methods in Robotics and Topology
Secured All India Rank 230 in JEE Advanced 2016 among 150,000 shortlisted candidates
Secured All India Rank 27 in JEE Mains 2016 among the 1.5 Million candidates
Qualified for INPhO and INAO conducted by Indian Association of Physics Teachers in 2016
Awarded KVPY 2014 fellowship, securing All India Rank 46 (out of a total of about 40,000 students)
Awarded National Talent Search Scholarship 2012 by National Council of Educational Research and Training
Secured 1st position in State level Essay Writing Competition by Govt. of Haryana in 2010
(Find details on these in my resume)
Deep Learning Controller for Autonomous Driving
Simulated evironment in Webots, tried Nvidia CNN end-to-end model and Conv-LSTM models, and their variations for turning on an intersection
Emphasis Selection for Written Text in Visual Media
Implemented 2 deep learning models using transformers models like BERT, RoBERTa, XLNet; stood 3rd in SemEval 2020 Task 10
GO to MIPS Compiler
Implemented a compiler in python for a subset of programs in GO language, targeting MIPS, using PLY framework with support for Dynamic Memory allocation and other advanced features
Probabilistic Control Strategy Synthesis
Studied the problem of designing a control strategy for a robot to maximize the probability of satisfaction of certain specifications formulated as LTL or PCTL formulas
Extended various functionalities of GemOS operating system; implemented system calls, added signal handlers and implemented scheduling using a round-robin scheme
Painter and Genre Classification
Used self-designed CNN, and feature extraction with transfer learning using VGG16 and ResNet50 models; gained a maximum accuracy of 75%
Deliver It App
Designed a community based delivery app using geolocation services on IONIC Framework; used Firebase Realtime Database and Leaflet maps
English Premier League
Implemented a database management system for a miniature scale model of EPL using LAMP stack
Fusion of Inertial Sensing IoT Devices
Implemented a fusion algorithm on Dual Foot-mounted Inertial Sensors’ data to reduce Systematic Heading drift
Designed and fabricated a Formula race car (F-18) for Formula Bharat; Secured 9th position in Design Event, 6th in Business Plan and 15th position among 55 teams at Formula Bharat 2018; As a part of Powertrain Subsystem, designed the sprocket on Solidworks, and simulated and optimized its design on ANSYS
Customizing web experiences using real-time user interaction data
Online user behavior is typically tracked using their clicks and analytics by enterprises. However, the current state of the user is not captured through such interactions.
We explored how user’s current web interactions combined with the content consumed can be leveraged to first understand the user cognitively and then create customized experiences while the user is still browsing in the current session.
We introduced 2 deep learning models, first one that predicts the user profile based on the interactions and the second that leverages this information to create a customized experience. The models take in real time data.
One of the major challenges was availability of no dataset for the problem statement.
Implementation and instrumentation of mechanisms to capture user information in real time from web interactions using JS libraries, which was a non-trivial task given the variance in different website designs
Hosted 2 types of surveys on Amazon Mechanical Turk (AMT) for data gathering using multiple websites
Led the conceptualization and data gathering for ‘customizing web experience’ part, along with implementation of models such as RBM and Autoencoders in PyTorch
Currently in the process of filing a patent and paper for the approach and work done in the field