Engineering owner of all mobile apps in PeopleStrong - Alt Worklife App, Zippi App (Messaging & Collaboration) and Jinie - PeopleStrong's AI driven enterprise SaaS HR chatbot

As a senior data scientist I was responsible for several strategic AI initiatives in the organization. Some of the key projects I was owned:Jinie Chatbot - Complete engineering ownership of PeopleStrong's AI powered enterprise SaaS HR chatbot. We scaled the chatbot from answering a few queries a day to several thousands of queries in a day thereby significantly reducing the number of support tickets raised across organizations. Our DAU was almost doubling every month and I was responsible for broadening the capbiliy of the chatbot in the following directions: automation, employee engagement, personalization, query fulfilment using hybrid models, conversational analytics, tracking KPIs and adding product delighters such as context aware conversations and smart nudges. We also revamped our NLU engine from using single to a hybrid model that combines open-source and cloud frameworks in an enterprise SaaS environment.Facial Attendance - Automated attendance marking through face recognition. Explored GANs (Generative Adversarial Networks), Google’s FaceNet (CNN based) and Google Vision. Created continuous training and deployment pipeline for daily new user registration.Job Recommendation - Recommending relevant jobs to candidates based on candidate-job features and interaction history of candidates with jobs. Used Genism’s Word2Vec API to create deep word embeddings of skills. Model was implemented using Bi-LSTM (Bidirectional Long Short-Term Memory Network) with attention mechanism. Research published in IEEE: https://ieeexplore.ieee.org/abstract/document/9073723Sentiment Analysis - Used texts exchanged on corporate messenger platform to analyse mood and sentiment of employees using DeepMoji

Responsible for leading a team that developed several models around candidate-job matchmaking. Some key projects I was responsible for:Building 2 distributed 24 x 7 crawlers. One that crawled 40+ social media websites to enrich candidate profile and the other that crawled the top 10 Indian job boards to generate skill, industry and job role embedding. Both used the same crawling technical framework & methodology. Few key problems we solved while fetching information from the public domain was to honor rate limits, devising counter measures to work around IP blacklisting and web browser fingerprinting Creating hybrid AI models for performing candidate - job matchmaking using data crawled from the public domain

Complete engineering owner of the automated code evaluation engine and DevOps / Infra - this was a mission critical 24 x 7 business area. Coding assessments scaled to a magnitude of over 100k assessments in a day and during this period we also signed up key accounts such as Microsoft and we managed their complete pan India fresher hiring (for Software Engineers). I also had the full engineering ownership of the mobile application development teamKey Projects:Conceptualized and created end-to-end specification of the code evaluation engine product, and managed teams that collectively delivered coding assessments to key clients like Microsoft, Wipro, etc. The product was built to be horizontally scalable and capable of evaluating 1800 submissions per minute per instance. Humanized few aspects of the evaluation technique. Came up with intuitive product ideas such as automatic error detection, static analysis for mapping candidate attributes form raw code without compiling and empirical estimation of cyclomatic complexityDelivered Android App for conducting assessments both offline and on the go. Offline app was needed to conduct assessments in remote areas where network connectivity is poor. Developed methodologies to sync assessments when network connectivity is available but intermittent. Solved for challenges such as syncing on SD Card to continue test taking process if devices go bad during the assessment processManaged DevOps and was responsible for setting up end-to-end High Availability (HA) on current infrastructure. Also automated several sync and failover jobs.

I worked on the engineering team that developed assessments and as a Senior Software Engineer some of my key responsibilities were:Owning the complete product development life cycle of the code evaluation engine right from conceptualizing new items to development, code reviews & deploymentOwned the complete DevOps for the code evaluation engine - did capacity planning and infrastrucure sizing to provide a 27 x 7 operating environment for the productContributed to novel product ideas and lead the RnD of the product to make it better from the competition at that timeStarted building & leading a new mobile development team. This was a strategic initiative to launch our assessment app on mobile platforms

I worked on the engineering team that developed assessments. We built the automated code evaluation engine from scratch. Some of my key responsibilities were:Writing code in C++ to develop the automated code grader and the containerized (jailed) environment for executing candidate's codeUse Linux kernel level hooks to ensure candidate's code run in a secure way and under time and memory limits.Writing code in C++ that automated grading of code on several qualitative & quantitative parameters such as warnings & errors, code runtime, memory usage, cyclomatic complexity, basic test cases, edge test cases and performance test casesTuned the grader to evaluate approximately 1000 submissions in under a minute

Contributed to the NextGen product in the following ways:Wrote code in C/C++ in the MAP and TCAP protocol stack layersImplemented several call flows from scratch with minimal supervisionWrote custom tools to detect hard to find memory leaks in code that are only revealed at a high TPS and go undetected with conventional tools - valgrind / callgrindTook part in deep dive technical discussions with customers to explain product architecture after delivery. Did sessions with Mahindra Comviva and C-DOTIndependently performed POCs for discovering new solutioning ideas for custom requriements

Contributed to creating a GUI based configuration utility to configure Aricent's NextGen product that was deployed in Mahindra Comviva. Some of the activities I performed were:Wrote code in JavaWrote JUnit test cases for automating testing and for use in regression