Work Experience

Dell Technologies - Office of the CTO (OCTO)

May 2025 - Aug 2025 | Austin, Texas, USA

Architected autonomous resource allocation system using 11 specialized AI agents with 57 tools, improving GPU utilization from 8→40%, achieving ∼25% cost reduction and 35-40% better decision quality
Deployed production system on real PowerEdge server fleets, processing 1000+ concurrent workloads with 89% cost efficiency, 91% success rate, and 26.5% improvement in decision quality over Kubernetes/SLURM schedulers
Built algorithm lifecycle management system with 4 AI agents enabling autonomous selection, extraction, validation, and zero-downtime replacement of production algorithms from academic papers via Semantic Scholar/arXiv APIs
Submitted 4 patents; Published internal paper OCTO-11136: Towards an Agentic Approach to Autonomous Resource Allocation

Aug 2023 - Jul 2024 | Hyderabad, Telangana, India

Developed DDS-GPT, a RAG-based tool using Google flan-t5-large LLM and instructor-xl embeddings that utilizes Dell Design System (DDS) documentation to generate code snippets for UI components, saving UI developer’s manual efforts
Automated the entire Product Experience Platform (PXP) CDO metrics dashboard, saving ∼ 4 days of manual effort per sprint for every product manager in Dell’s eCommerce Org
Worked with cross-functional teams to improve a Gedis class predictor, resulting in a 12% increase in SKU accuracy
Engineered a customer support bot using a Dell-hosted fine-tuned Llama2-70B chat LLM on Dell’s Jira Data to dynamically route incident tickets to appropriate SMEs, reducing 40% of SLA times
Led the adoption of MLOps within Dell’s ecommerce Org, enhancing ML models monitoring and automating retraining processes
Built a dynamic forecasting system for the Product Data Service Layer team to alert 59 clients, saving infra cost by ∼ 40%

Feb 2023 - May 2023 | Hyderabad, Telangana, India

Developed Machine Learning capabilities to analyze and extract precise error patterns within daily failed Job Logs for the Enterprise Business Intelligence (EBI) team, contributing to enhanced data-driven insights and decision-making processes
Fine-tuned the RobertaBert using Splunk Tables’ generated Error Logs data, achieving an F1-score of 98.03% with an inference time of 22 min 36s for 10 million records
Devised and implemented a combined model pipeline integrating Decision Trees, Naive Bayes, and XGBoost, strategically dividing data based on the best-performing model for each data pattern cluster, resulting in a significant reduction of inference time to 3 min 41s for 10 million records
Designed end-to-end ML ingestion pipeline to systematically monitor and retrain the model whenever failure jobs deviated from the existing cluster scope, ensuring continuous model relevance and accuracy
Deployed and Operationalized the model on Pivotal Cloud Foundry (PCF), optimizing processing speed to 18ms per record while efficiently handling batches of 1000 records in 3.2s

Jun 2022 - Sep 2022 | Hyderabad, Telangana, India

Worked with the Enterprise Business Intelligence (EBI) team to automate and enhance data reliability processes, streamlining data workflows and ensuring data integrity
Developed a feature for D’Owl tool to remove feature-level duplicates from records, enhancing data quality and reducing redundancy
Optimized the D’owl codebase, reducing processing time from an initial 7.68 milliseconds to an astonishing 997 microseconds for 1 record
Explored NLP capabilities to extract valuable information from emojis rather than simply removing them from records

Sep 2022 - Jan 2023 | Bengaluru, Karnataka, India

Improved Swiggy’s image similarity determination model, developing a supervised model on top of Barlow Twins’ image embeddings
Achieved an accuracy boost of 4.51% on the validation set and 12-13% on two test sets compared to the previous approach
Engineered a streamlined Computer Vision (CV) pipeline using Streamlit for Swiggy Dine-out, seamlessly generating advertisement videos from static restaurant and food images
Automated the manual embeddings generation process for Swiggy Supr Daily reducing manual efforts
Collaborated on the Swiggy x IIT Jodhpur project, exploring object detection models for facial verification