Aug 2021 – Jan 2022 · 5 mos

• Built, optimized, and stabilized Config driven, self-serve Identity Resolution Data Platform (IDR)
• The IDR platform provides a single customer view and helps the brand reach its audiences via online and offline IDs
• The IDR platform hosts one of the largest identity graphs (~ 40B vertices, 25B edges), build using JanusGraph and backed by storage engine Scylla
• Build and streamline ingestions, consumption, backup & recovery, cost monitoring pipelines
• Reduced spark jobs failures by adding rate limiter in data platform
• Added Privacy compliant framework like GDPR in the platform
• Enabled OLAP queries on the native graph by exporting data to the deep storage area (GCS/S3) in the component form
• Built Single click backup and recovery solution for JanusGraph
• Added leader election process using zookeeper in control-plane service to enable special capabilities to leader instance
• Added Hikari in the control-plane service for managing connections
• Unification of Identity Platform
• Evaluated Scylla's performance. Evaluated Materialised view, Lightweight transactions, bulk read-write using the spark-Cassandra connector.
• Unified the identity and profile store to provide a single source of truth for third party identity data
• Enabled CDC on graph ingestion pipeline to detect changes happening on graph structure. This reduces overall costs to our OLAP capabilities on the graph.
• JanusGraph:
• Graph of identifiers, connected with a bidirectional edge
• Used for Identity Resolution, finding the desired identifier for a given input identifier using DFS (or BFS)
• Modelled the graph into a KV Store for OLAP capabilities
• >> A utility to decode Janusgraph encoded datasets into human-readable datasets (Anatomy of edgestore table) - https://gist.github.com/bsnuag/7cce8a99171ec798f7d4e2f1437399d9

Mar 2019 – Sep 2021 · 2 yrs 6 mos

Feb 2017 – Aug 2018 · 1 yr 6 mos · Bengaluru Area, India

• Matching Engine:
• It enables clients to identify similar products from a competitor's inventory (Inventory is a collection of product information crawled from e-commerce pages)
• This engine powers the product suggestions being shown on e-commerce websites
• The matching process runs on product info (or product attributes) and connects the corresponding vertices in graph DB
• Graph:
• The products are represented as vertices and a match between two products is represented as an edge
• The product attributes are stored in ElasticSearch for text search
• The reporting service is powered by HBase for bulk exports
• >> Tools & Technologies Used: Java, Netty, Storm, Kafka, HBase, Elasticsearch, Titan (now Janus Graph)

May 2014 – Feb 2017 · 2 yrs 9 mos · Bengaluru, Karnataka, India

• Built and maintain customer info portal for Fidelity (Client)
• Worked primarily on backend services
• >> Tools & Technologies Used: Java, Hibernate, Spring MVC