Jan 2023 – Jan 2024 · 1 yr · Bengaluru, Karnataka, India

• Dev0 is a code intelligence platform to help developers effortlessly comprehend and troubleshoot microservice architectures. We eliminate ‘change-anxiety’ by tracking dependencies across codebases as we intercept data flow between services. Further, our git-like change tracking system lets you analyze all changes(code, config, network, infra) for a microservice in a single glass pane. This saves developers tens of hours per week and accelerates incident resolution.

Jan 2021 – Jan 2023 · 2 yrs

• Building an ETL framework to pump reference data across the entire Goldman Sachs Electronic Trading system using our in house Core Driven Architecture codenamed, Atlas.

Nov 2020 – Apr 2021 · 5 mos

• Spearheaded the NLP research in very early stages of startup which aims to add emotional intelligence to video calls.
• Trained deep learning models to extract action items, pain points and other important trackers from call transcripts.
• Finetuned pre-trained language models for the task of multi natural language inference to handle oov text in production. •Deployed models to segment call transcripts based on topic of discussion using Tensorflow’s universal language encoder.
• Formulated algorithms which build upon above signals and extract questions and answers exchanged during the call.

Aug 2020 – Apr 2021 · 8 mos

• OxKBC: Outcome Explanation for Factorization Based Knowledge Base Completion, accepted in AKBC 2020.
• CEAR: Cross-Entity Aware Reranker for Knowledge Base Completion, submitted in EMNLP 2021.
• Orchestrated a system to generate post-hoc explanations for predictions from factorization based KBC models.
• Defined human understandable explanation templates, trained a novel selection module which explains the given fact.
• Capitalized world knowledge stored in pre-trained models using a 2-stage architecture built on KBC embedding models.
• Achieved state of the art performance on standard dataset FB15k-237 as well as a novel open world dataset-OLPBENCH.

May 2020 – Jul 2020 · 2 mos · Bengaluru, Karnataka, India

May 2019 – Jul 2019 · 2 mos

• Idea is to load webpages faster by caching static content of a website and only requesting dynamic details.
• Build a system that can patch these dynamic responses in your static page to generate the complete html.
• Analyse Reddit’s and Sharelatex's backend to develop a method for automatically generating the static content.

May 2018 – Jul 2018 · 2 mos · Greater Delhi Area

• Contributed to Research Paper, titled Embedded CNN-Based Vehicle Classification and Counting in NonLaned
• Road Traffic, accepted in ICTD 2019 as lead author.
• Ran object detection models on Raspberry Pi 3B, Intel Movidius stick, Jetson TX2 and Android Devices.
• Motivation: in-situ processing of traffic feeds would cut the cost of streaming the videos to GPU servers.
• Determined the mAP accuracy, benchmark scores and energy values on the KITTI dataset.
• Modified caffe source to build an implementation for python that could profile all layers of object detection models.
• This project is under review for Summer Undergraduate Research Award.