1. Migrated backend from woocommerce to self-managed Flask based webserver, helping scale from 100 orders to peak of more than 5000 orders per hour.2. Reduced p99 server side latencies from 1000ms to <= 200$ms for read heavy endpoints:a. Introduced LRU and leveldb over redis, reducing network calls.b. Removed usage of redis blocking commands like keys.c. Migrated wishlist and cart to dynamodb, removing all but create order api dependencies on RDS.d. Added support for horizontal scaling of caching (redis) and search (solr) infra, leading to reduced latency and increased read throughput.3. Revamped inventory sync pipeline to reduce inventory info lag from 24 hours to <= 5 minute, enabling NDD and SDD delivery4. Completely owned Analytics and Observability Infra by self hosting Clickhouse instance.5. Migrated integrations with all external logistic and payment partners from pull-based to hybrid push-pull based, leading to reduction of p99 for item tracking from 6 hours to <= 5 minute.6. Worked with datascience team to improve search query understanding, bringing down null search from 18% to 6%, and increase in CTR of search results by 8%.7. Managed team of 15 members, playing an active role in hiring and onboarding.

1. Designed and developed the initial data model for supporting different personas (candidate, hiring companies, career supporter and exchange sponsor) of Career Exchange.2. Developed a pipeline to support recurring ingestion of 2M+ jobs daily from National Labour Exchange; extended the pipeline to support recurring ingestion of courses. Added support for large files by introducing multi thread download from sftp server reducing time by 50%, added monitoring and alerting for better observability.3. Improved retrieval pipelines for jobs for exchange, primarily focused on blue collar jobs, resulting in 10% jump in CTR.4. Enhanced analytics pipeline to power dashboards for exchange use-cases.

CMSC828U Machine Learning with Guarantees and AnalysesCMSC216 Introduction to Computer SystemsCMSC250 Discrete Structures

Content discovery using Knowledge graph embeddings.

Worked on developing a vector image editing IOS app based on Adobe Illustrator, in particular working with the pen, direct selection, selection and the snapping tool. Worked on developing the Adobe's text shaping library for IOS, Win and Mac.

Working on enhancing WRServices, Adobe's text shaping library used by most of the Adobe products like Photoshop, Illustrator and InDesign to handle the text shaping of complex scripts, by integrating it with Harfbuzz, the popular open-source text shaping engine, allowing for shaping of more than 40 scripts.

Advisor: Prof Venkatesh RamanWorked on designing an efficient Fixed Parameter Tractable Algorithm for finding the Partial Vertex Cover in Bipartite Graphs and proving the lower bounds for the algorithm. Also explored the vertex cover and the feedback vertex set in tournaments and the state-of-the-art FPT algorithms available for them. Gained experience in the different standard reduction techniques to prove the fixed-parameter tractability of the various graph-theoretic problems.

The project involved study of Computational Linguistics to build a system that assigns semantic labels to edits between versions of a document. The built system encompasses algorithms to align the sentences between the versions; identify and classify them into Paraphrasing and Factual changes, and their subcategories. The algorithm also provides a ranking module that, based on user feedback and using Extractive Summarization Techniques, learns a regression model that is used for ranking the various types of edits for easier reviewing.The work was accepted at EMNLP 2017 as a short paper, and a patent titled "Classifying and ranking changes between document versions" was accepted by the US Patent office, Publication No: US2018/0285326 A1, Issued on Oct 4, 2018.

Worked on Algebraic Number Theory problems under the guidance of Prof Dipendra Prasad

Studied the works of Brandes et. al. and Chehreghani on techniques to compute Betweenness Centrality and designed a non-uniform sampling technique of nodes in a graph to compute it. Implemented the Brandes algorithm in C++ along with the proposed technique and ran experiments on Erdos Renyi graphs, Barabasi Albert graphs and Stanford SNAP dataset. The proposed technique significantly outperformed the existing techniques in most of the experiments.The work was accepted as a short paper "An efficient estimation of a nodes betweenness" In Complex Networks VI (pp. 111-121), Springer, Cham, 2015 and as a journal paper in Social Network Analysis and Mining, 2018.