Apr 2023 – Nov 2025 · 2 yrs 7 mos

• Redesigned ShareChat’s real-time ML feature platform(TARDIS) by migrating from Flink SQL to the Flink DataStream API,
• Replaced Flink SQL abstraction to write stable operator IDs, enabling seamless savepoint-based job upgrades and reduced failure risk.
• Achieved ~20% CPU reduction by shuffling data as bytes across TaskManagers, enabling parallelized deserialization and better CPU utilization.
• Enhanced production observability and debugging with custom metrics.
• Built GCS filesystem connector for TARDIS, Added support for reading filesystem-based sources in TARDIS by building a GCS connector, enabling ingestion of diverse batch data similar to Kafka-based pipelines; unblocked migration from legacy systems (UFS, Bigtable) and significantly reduced operational overhead.
• Contributed to building the real-time training pipeline, focusing on batch-side processing to ensure high-quality feature and target data for ML model training.
• Defined protobuf schema and implemented feature logging, enabling structured logging of features (as tf.Example) into Databricks delta tables.
• Developed a robust feature health monitoring pipeline, including Spark jobs to flatten serialized user/post features and BigQuery-based health metrics (nulls, match rate, etc.) to ensure training data quality.
• Implemented an hourly batch joiner Spark job to merge features and targets into TFRecord format, directly consumed by ML training jobs; now adopted by multiple teams.
• Created a Spark job for validating target data quality, detecting nulls and mismatches across BigQuery and DB sources to build confidence in downstream training.
• Designed and implemented cross-feature support in ShareChat’s Flink-based feature platform (TARDIS) by adding ScyllaDB lookup join capabilities; supported use cases like user-topic feature joins and optimized pipelines for peak traffic, leading to 40% lower compute usage and ₹28.2L/month cloud savings.

Apr 2022 – Apr 2023 · 1 yr

• Developed and scaled Moj’s Boosting Service — a Go-based microservice powering real-time content prioritization for feed ranking with sub-second latency across 15+ Indian languages.
• Original Creator Boosting: Integrated Creator Grade Service (CGS), replaced legacy pipelines, and launched A/B variants to improve OC view share.
• Creator Diversity Boosting: Fetched creator counters via Tardis, implemented fixed boost rules, and shipped variants to enhance diversity in feeds.
• Min View Boosting: Prioritized underexposed posts using hourly view counts and exposure thresholds to ensure baseline visibility.

Jun 2021 – Mar 2022 · 9 mos

• Led development and A/B testing of affinity-based candidate generators (Creator, AudioID, Topic, Lens/Cluster), enabling feed personalization through entity-level affinity signals. Collaborated with product and infra teams to iterate rapidly, reduced latency through code optimizations, and shipped a winning variant with a 0.25% retention gain.
• Designed and built a promoted post pipeline to support business use cases like livestream promotion and content boosting. The pipeline ingests post IDs with view counts and TTL, computes priority dynamically, and writes to Redis for consumption by feed candidate generators, enabling controlled injection of promoted content into user feeds.

Mar 2021 – Jun 2021 · 3 mos

• Built a Pub/Sub pipeline to update real-time status in Spanner and write static features to Bigtable, integrating with proxy services for downstream feature serving.
• Optimized BigQuery performance by clustering key tables (video_play, feed_events) via schema and Dataflow updates, reducing query costs with zero ingestion downtime.

Nov 2020 – Mar 2021 · 4 mos · India

• Managing a team of Problem reviewers and Problem setters to develop Problems in the domain of Data Structures, Algorithms and Software Engineering for the Project of Interview Experience and also managing and leveraging the quality of content.

Sep 2020 – Oct 2020 · 1 mo · India

• Managing Problem Setters, reviewing problems and approaches developed by them in the domain of Data Structures, Algorithms and Software Engineering.

Jul 2020 – Aug 2020 · 1 mo · India

• Working on Problem Setting, Testing and writing approaches in the domain of Data Structures, Algorithms and Software Engineering.

Aug 2020 – Nov 2020 · 3 mos

• Improved the Codes, Test Cases, Problem Description and Constraint
• of around 100 Data Structure and Algorithmic problems.

Jun 2020 – Feb 2021 · 8 mos

• Use to create different algorithmic problem like greedy , DP ,number theory ,bit manipulation etc for different contests at HackerEarth.

Jan 2020 – Apr 2020 · 3 mos

• Working as a student developer at Crio.Do during Crio Launch 2020 has been great with lot of challenges. Learnt a lot about industry expectations and the challenges faced while working on real-time projects .Learned from basic to advance concepts of Linux Commands, OS Concepts, Networking Concepts, REST API, Jackson, Interfaces, Code Refactoring, Gradle, Problem solving and a lot more.
• Here is the link to portfolio of projects : https://criodo.github.io/Crio-Launch-Feb-2020-mdazmat9/

Jul 2019 – Nov 2019 · 4 mos

• My job as TA’s at coding ninjas were to clear the doubts,debug their
• code,give them hints etc to the students who took competitive
• programming course and it was paid internship for 4 months
• And solved and debugged around 350 doubts with Average rating 4.6.

Mar 2018 – May 2018 · 2 mos