Apr 2024 – Jun 2025 · 1 yr 2 mos · London Area, United Kingdom · On-site

• Optimized billion-user-scale Ads ranking models by designing and implementing new features capturing Advertiser and User purchase price affinities, boosting advertiser ROAS.
• Drove feature engineering initiatives for precise ad targeting across Retrieval, Pre-Ranking, and Ranking stages.
• Built scalable feature engineering tools, reducing iteration cycles by 30% to enhance team productivity.

Jul 2021 – Mar 2024 · 2 yrs 8 mos

• Led cross-functional teams to develop and deploy advanced ML models for competency identification and ranking, across VMock’s products. Designed & implemented scalable, reliable ML delivery system using Docker/Kubernetes(EKS, SageMaker) on AWS Cloud
• Resume Bullet Generation with Fine-Tuned LLMs
• Fine-tuned Microsoft’s Phi-2 (2.7B) using LoRA adapters to generate resume bullets from skills, industry, and role inputs
• Optimized training with Mixed Precision and DeepSpeed, reducing memory usage and training time
• Skills Recommendation Engine
• Retrieval Stage: Implemented three candidate generators: Sentence-BERT for skill attributes, Matrix Factorization to capture co-occurrence, and Node2Vec + GraphSage for Skills Ontology, achieving 92% recall in retrieving the top 300 skills from 40,000.
• Ranking Stage: Employed GBDT-based LGBMRanker to predict top-k skills, attaining 87% nDCG@10 on the test dataset.
• Skills Detection in Resume Bullets
• Developed a multi-label classification model to accurately identify skills from 40,000 distinct labels in resume bullets.
• Utilized Deep Siamese BERT with Contrastive Loss, achieving an 80% recall rate. Employed bagging to address class bias and imbalance.
• Skills Rating: Evaluating candidates' key skills in their resumes – Utilized Skills Ontology, Graph Algorithms (Dijkstra & HITS-Rank), and resume factors to effectively assess the relevance of candidate skills.

Jul 2018 – Jul 2021 · 3 yrs

May 2017 – Jul 2017 · 2 mos · Gurgaon, India

• Developed a standalone software in Python to extract Deeplinks and Attribution SDK’s from apks.
• Created UI and server in Django with Mongo Database for multiple user access.
• Implemented "ProGuard" on apps to increase security against cloning and piracy.

Apr 2016 – Mar 2017 · 11 mos · Aravali Hostel, IIT Delhi

• Led a team of 11 representatives to secure 4 podium finishes in cultural events.
• Organized Inter-College Socials and fundraised with LSR College for women in 2016.
• Conceptualized and executed Intra Hostel Festival, Exuberance, featuring 9 cultural events over 3 days in 2016.