Jun 2021 – Feb 2022 · 8 mos · East Palo Alto, California, United States

• Developed a Rust/C++ SDK for sandboxed, serverless execution of customer third-party extensions in PostgreSQL, without need for manual deployment verification by DB administrators. Responsible for :
• Provisioning multi-AZ databases across a variety of EC2 instance types and EBS volume types
• Automated backups and cross-region streaming with multiple read replicas
• Auto engine version upgrades, patching and failovers aided by disaster recovery using S3
• Support for custom DB parameters for better query workload performance

May 2020 – Aug 2020 · 3 mos · East Palo Alto, California, United States

• Integrated an open-source extension PL/Container into PostgreSQL 12 and 13 to facilitate sandboxed execution of untrusted functions & stored procedures inside Docker containers to prevent customer programs from crashing the database server and enforce finer access privileges
• Performed extensive benchmarking with a study of shared memory segments, Unix sockets and grpc channels to identify bottlenecks for maintaining average latency at par with extensions running locally
• Developed a new prototype leveraging a single container across all customer sessions, leading to a 63% reduction in memory usage while scaling to thousands of customer connections
• Added runtime support for Go language, and created 3 separate extensions for each of R, Python and Go utilizing separate containers for better isolation

Aug 2020 – May 2021 · 9 mos · Madison, Wisconsin, United States

• Log-Structured Merge (LSM) based key-value stores have become so popular today that they are used as backend for relational database abstractions like TiDB. They use indexes for faster data lookup whose memory overhead increases with database size, leaving lesser memory for caching data blocks. Recent studies also show that application throughput can be compromised by internal LSM tree operations that periodically write data to disk.
• Observed 2.24-2.34X improvement in client write throughput for industrial write bursty workloads by scheduling background memtable flushes and compactions during idle periods or periods of very few writes to the database
• Employed Learned Indexes to train a model to learn offsets from last keys of data blocks inside SSTables to reduce lookup time from O(logn) to O(1)
• Obtained a 53% reduction in indexing memory footprint over traditional indexes with <5% increase in read latency using both Fuzzy and Greedy Piecewise Linear Regression in RocksDB
• Guides : Prof. Remzi and Andrea Arpaci-Dusseau

Jan 2020 – May 2020 · 4 mos · Madison, Wisconsin Area

• CS 559 : Computer Graphics
• Instructor : Prof. Michael Gleicher

Aug 2019 – Dec 2019 · 4 mos · Madison, Wisconsin Area

• CS 559 : Computer Graphics
• Instructor : Florian Heimerl

May 2019 – Aug 2019 · 3 mos · Greater Seattle Area

• Worked with different L4 and L7 proxies to demarcate functionalities for offloading to the SmartNICs. Carried out benchmarking experiments across Nginx, HAproxy and Envoy with SSL to determine feasibility of checksum offloading and scalability. Performed a detailed study of Envoy source code to determine scope for distribution of core functionalities between host and hardware for offloading
• Guide : Prof. Arvind Krishnamurthy

Jan 2019 – May 2019 · 4 mos

• Served as a Teaching Assistant for a Freshman year course, Computer Programming and Utilization.
• Instructor : Prof. Ganesh Ramakrishnan
• Responsible for evaluation of labs for a batch of 450+ students
• Set up weekly lab statements on bodhitree : IITB's online classroom learning initiative

Jul 2018 – Nov 2018 · 4 mos

• Served as a Teaching Assistant for a Third Year Computer Science Core Course, Computer Architecture
• Instructor : Prof. Bernard Menezes
• Entrusted with evaluating and grading labs for a batch of 120+ students
• Assisted in setting up weekly lab statements and semester project proposals

Jun 2018 – Jun 2019 · 1 yr

• Represented the department of Computer Science and Engineering in the institute placement team
• Assisted in the 2019 annual placements by conducting preparatory tests and personal interviews
• Verified and scrutinized resumes of over 50+ students across all departments

May 2018 – Jul 2018 · 2 mos · Bengaluru Area, India

• Adapted Facebook AI Research’s Convolutional seq2seq model for translation to characteristic-driven text generation on pytorch, using CNNs for computing encoder and decoder states
• Modified beam search paradigm for enhancing formality, descriptiveness and vocabulary tailored generation; attention
• layers for topic-tuned summaries and token-based learning for length based summarization
• Incorporated a Reinforcement Learning term in loss function and achieved a 6.4% increase in ROUGE scores
• Guide : Dr. Balaji Vasan Srinivasan

Nov 2017 – Dec 2017 · 1 mo · Mumbai

• Involved in the assessment of retail customer database for appographic targeting through contextual marketing and developing focused strategies for the same
• Leveraged Natural Language Processing techniques for customer segmentation and retail-category-affinity estimation in python
• Implemented probabilistic graphical model based recommendation engine with contributions to the pgmpy github repository
• Created a new query language for internal database query system on neo4j utilising parse trees

May 2017 – Jul 2017 · 2 mos · Bengaluru Area, India

• Designed a chatbot for customer care services to resolve queries pertaining to various consumer products
• Facilitated the parsing of user manuals for construction of chatbot knowledge base with the addition of an AIML engine
• Built the Semantic Engine for contextual semantic matching to find the most relevant topic leveraging Latent Semantic Indexing an Latent Dirichlet Allocation in python gensim with enhanced synonym recognition
• Built a Sentiment Classifier for tracking customer satisfaction in live chat sessions based on a multiclass, multilabel emotion classifier with a backend sentiment score plotter
• Facilitated the building of knowledge representations using ontologies from XML, CSV sources using Stanford’s Protege Library with an interface to fetch data responses from the same using SPARQL queries

Jan 2017 – May 2017 · 4 mos · Mumbai Area, India

• Served as a Teaching Assistant for a Freshman Year Course, Introduction to Biology
• Instructors : Prof. Ambarish Kunwar, Prof. Rahul Purwar, Prof. Sanjeev Srivastava and Prof. Prakriti Tayalia
• Entrusted with evaluating and grading papers for a batch of 450+ students
• Conducted weekly tutorials and helping sessions for clearing doubts on a one-on-one basis

Dec 2016 – Jan 2017 · 1 mo · Mumbai Area, India

• Worked on the automated timetable generation algorithm from university slotting data :-
• Designed the backend algorithm for generating timetable for institutes based on a set of constraints encompassing room compatibility, professor availability and course working hours
• Implemented genetic algorithms in Java for obtaining the best-fit optimal timetable, mimicking biological evolution
• Added sync to MySQL database and displayed the same via PHP, hosted on server for tracking availability and managing ad-hoc modifications to the timetable schedule

Apr 2016 – Aug 2016 · 4 mos

• Working in a team of 24 towards conceptualization and successfully organizing Mood Indigo 2016.
• Pursuing the marketing budget required for Mood Indigo 2016 through corporate sponsorship.
• This includes marketing activities and integration(s) tailor-made for each brand during the event so as to mutually add value to both Mood Indigo and our partner.
• Conceptualizing and Executing brand integration strategies having Indian youth as the target group for global and national corporations.
• Primary focus on promoting brand Mood Indigo by developing synergic relations with corporates through Relationship Marketing.