In-House Hybrid Search – Transitioned ZEE5’s search from a vendor, Sensara to in-house, which led to a massive decrease in cost incurred by Search from 45 million to 5 million annually
Ideated and developed in-house hybrid search system incorporating both full-text and LLM-powered search which would provide users with a conversational search experience, along with query suggestions
Full-text search handles Title Search and Episodic Search which is a specially curated feature for OTT platforms where users can search based on episode, specific day, genre, and star cast enhancing content discovery
Performed load testing on the service and noticed latency being too high for broad queries. Implemented query caching using Redis to store the query embeddings, thereby avoiding repeated calls to embedding generating API and reducing the latency
Service has shown that it is capable of handling 300 TPS on average and 600 TPS during peak hours with P99 latency being approximately 150ms
Created a data pipeline with Google Cloud Functions, that will consume any new data published, perform certain transformations, and update SOLR in line with business rules and data security standards
Transition to in-house hybrid search system has achieved more reliable search results than the external vendor resulting which the Search CTR increasing from 79% to 91%
Generative-AI Search – Developed LLM-based search to handle broader queries by providing a set of similar rails that would match the user's query
Leveraged SOLR for vector database, where user query is converted to an embedding to retrieve top ’N’ similar rails
Built a data pipeline with Google Cloud Functions, that gets triggered when the ML team updates rail and the content mapping in Google Cloud Storage, thereby updating the same changes in SOLR
The feature had been rolled out for 100% of the iOS users and noted a significant increase in the Search CTR from 79% to 86%
Associate Software Development Engineer
Zee Entertainment Enterprises Limited
07.2022 - 11.2023
Developed APIs to create, update, and delete rules to ensure recommended contents are effectively shown to users and meet the business requirements simultaneously
This would enable the plug-and-play of different ML models based on certain criteria, or to pin newly released content at the top of the rail for a certain period
Designed and developed a dashboard that would perform all the operations mentioned above, making it easier for team members and product managers to see how the rail would look and perform when it goes live
Built a feature for the dashboard, where the team can curate their static rails and get an estimate of Click Through Rate (CTR) when it goes live
These rails can be switched on or off on a flag basis.
Software Development Engineering Intern
Zee Entertainment Enterprises Limited
01.2022 - 07.2022
Contributed to increasing the unit test code coverage for Recommendation systems from 33% to 85%
Developed a schema validation service to effectively handle and log any change in the request body sent by the front-end for the Recommendation service using NodeJS
Project Intern
Sasken Technologies Limited
10.2021 - 12.2021
Developed an intent-based code generator engine to generate the code in pyTest format by mapping the functions
to the respective intents.
Summer Intern
Indegene
06.2020 - 07.2020
Built an AI-based chat-bot that provides the employees with the answers to their queries
The bot was able to
handle the fallback questions and was successful in classifying multiple intents.
Education
Bachelor of Technology - Computer Science & Engineering
PES Univeristy
Bengaluru, KA
2022
Skills
Languages: Java, NodeJS, Python
Backend Development: Spring Boot, SOLR, Kafka, Cassandra, MySQL, Redis
Miscellaneous: Google Cloud Platform (GCP), Amazon AWS, Search, Recommendations