• Designed and launched the first neural-network-based enforcement classifier on WhatsApp (2.5B+ MAU), targeting scam

actors and reducing scammer-engaged conversations by 18%.

• Independently identified and pursued the initiative to curb Victim-Initiated Scams—contributing to >50% of all scams on

the platform—by researching, prototyping, and validating enforcement models across two new operational contexts

(disconnect events and wa.me entry points).

• Owned the full initiative lifecycle end-to-end: conducted impact analysis, trained classifiers operating on 300M+

disconnect events and 40M+ wa.me events daily, designed and executed 3-week A/B experiments, led experiment result

evaluation with the team, and launched to production—delivering an 11% reduction in the SEC topline metric, eliminating

~3M Scam Engaged Conversations on a 7-day rolling average.

• Collaborated with the SMB (Small & Medium Business) team to extend enforcement models to business accounts,

broadening the integrity coverage surface across WhatsApp.

Working on advanced AI Agent systems leveraging Large Language Models (LLMs) and

multimodal inputs. Focused on applied research and ML system design.

• - Developed an automated problem-solution extractor and multi level ranking algorithm using POS tagging, SVD based QoS score, domain specific patterns and keywords. The solution now acts as a one stop solution for the client vendors to access multiple tools, get virtual SME support, new onboard's virtual guide and build on the knowledge repository.

• - Designed deep learning based encoder-decoder architecture for the solution generator block with ReLU boosted LSTM as building blocks, trained on historical tickets and deployed as a web API using Flask.

• - Worked on document search API development for the design documents such as HLD,LLD, acceptance criteria docs, wikis and Ranking using Tika and tf-idf matching that improved the Defect Retest Efficiency by 17 percent.

• - Developed the feedback system for the predicted solutions with cloudant db.

• - Teaching Assistantship, Machine Intelligence and Learning, (July 2018 - Dec 2018) where jointly conducted tutorial sessions on Optimisation Problems, SVMs, CNNs, RNNs and GANs for the under-grads of IITD.