May 2024 – Dec 2025 · 1 yr 7 mos · Noida, Uttar Pradesh, India · Hybrid

• Enhancing Naukri.com's search capabilities to drive innovation and connect recruiters with the right candidates with Agentic search using RAG, generative retrieval, dynamic facet generation, hyper-personalization, guided discovery and provide the best search experience to the user.

Aug 2022 – May 2024 · 1 yr 9 mos · Noida, Uttar Pradesh, India · Hybrid

• Part of core search team at Infoedge's naukri.com
• As a core member of the search team at Infoedge, I am responsible for developing and optimizing algorithms to maximize CTRs and other key business metrics such as clicks and downloads. My contributions included:
• 1. Core search architecture revamp: Worked on all components of the search system, from base ranker to re-ranker and personalization.
• 2. Base Ranker Optimization:
• Enhanced the base ranker on top of Elasticsearch to maximize recall.
• Developed a novel optimization algorithm and an AI verification framework, resulting in an 8% increase in business metrics.
• Experimented with vector DBs like Milvus.
• 3. Reranker Improvement:
• Trained a LAMBDAmart and Random Forest model on base ranker top-500 results to maximize precision, leading to a 6% increase in CTR.
• Used aerospike and query caching to limit the online retrieval time to 80ms.
• Implemented a cross-encoder re-ranker setup for optimizing ranks.
• 4. Embedding Training:
• Developed query and document embeddings using FastText and BERT encoders.
• Implemented a bi-encoder architecture optimized with contrastive loss for contextual embeddings.
• 5. Cross-functional Collaboration:
• Actively coordinated with the tech search, product & QA teams for performance testing (PT), leadership PAT, A/B testing, and go-live processes.
• 6. Live Monitoring:
• Created a live monitoring pipeline using Spark to track key metrics in real time.

Nov 2021 – Aug 2022 · 9 mos

• Churn Prediction: Aims to reduce the customer churn by 25% by Q3, 22
• E2E owner of the competitor impact as a feature on churn. Designed ready to use geo-spatial algorithms/packages to identify coverage index of Airtel VS Competitors using MongoDB.
• Leveraged web-scraping tools like selenium, beautiful soup & scrapy to pull property centric data like units, floors, price etc. from sites like quickr, 99acres, and google maps API for geolocation.
• Wrangled through enormous IMSI level data of ∼300M customers, ∼10B geographical coverage data and provided analysis / insights for feature development.
• Used customer segmentation techniques to arrive at a CEI (Customer experience index) using KPIs like coverage, throughput, MoUs, customer complaints etc.
• ASON-Duct Prediction: Prediction of atmospheric duct in TD-LTE networks
• Atmospheric duct is known to have detrimental impact in telecom on user experience and aggravated complaints.
• This module aimed for customer complaint reduction by 40% and UL RSSI by 20% and triggering ∼2M network changes per day.
• Designed a rasterization and interpolation-based algorithm to predict duct a day in advance, and backend APIs to capture the impact and outputs on-the-go using Flask.
• Developed and deployed the regression model to production, seamless implementation of data pipelines of weather APIs, network KPIs, OSS dumps on top of HDFS cluster using PySpark.
• Crafted an NLP based Outlook BOT for ad-hoc reports availability to clients in no time

Jun 2020 – Nov 2021 · 1 yr 5 mos

• Site Profitability Module: Ensure 100K revenue for all new site deployments
• Worked on to identify new site locations and azimuth recommendation in rural India using deep learning semantic image segmentation.
• Created a model to predict the approximated revenue the site will draw after launch.
• ◦ Used CNN to map the clutter / population using google maps satellite imagery and feed to a regression
• model on top of it to predict revenue.
• Performed semantic segmentation using FC-Densenets with varying hidden layers depending on geography and IOU as a metric.

Jun 2019 – May 2020 · 11 mos

• Developed an algorithm to cover villages PAN India using adequate number of sites to ensure business profitability. Analysed renewable energy solutions to power up the site.
• Developed a broadband churning model based on Xgboost and Gaussian SVM. Achieved an recall of 67% to predict the churning of customer using broadband.
• Increased the sales in Serve to Sell (S2S) approach by 3X. Used the applications of web scraping and NLP to analyse the data and captured data points from multiple LOBs.
• Understood MANO architecture and the application of NFV and SDN to improve network functioning.

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

• Awarded Best Intern of batch 2018 and PPO.
• Key highlights:
• Developed a standalone software "Bharti Automation Tool" on Python and PyQT. The software consists of 5 modules along with an add-on utility.
• 1. Network Analyzer - Module capable of creating trends of KPIs with respect to days for all technologies 2G, 3G and 4G with a click of a button!
• Time reduction - 540 times.
• 2. SCFT simulator - Module can verify 500 -600 SCFT reports in just 10 seconds using RBF kernelized SVM and OCR for the implementation.
• Time reduction - 1000 times.
• 3. SACFA punching Bot - Completely automated the process flow and reduced tons of manual punching of data using selenium and beautiful soup.
• 4. New Sites village finder
• Add-on utility - Developed a automatic digital signature bot capable of signing PDFs present in a dump.