Experience

MBZUAI · Supervisor: Dr. Nils Lukas

• Built ingestion/cleaning pipelines for Adobe and NICE; extracted personas and planned journeys; unified these into a single retrieval layer for the agent.
• Designed structured response schemas (single/multi-flight) with “flight transparency,” and extended planning/LLM logic to support all flight options on request.
• Implemented real-time audio capability for the Etihad chatbot using ChatGPT's streaming API, enabling full-duplex voice interactions.
• Integrated critical APIs (incl. return-journey), enabled targeted supervised fine-tuning, parallelized conversational state for latency, and added bi-directional audio with the same policies.
• Designed and simulated a UserLM to emulate user behavior patterns, supporting training, evaluation, and stress-testing of the conversational system.

MBZUAI · Supervisor: Dr. Nils Lukas

• Strengthening LLM-agent pipelines against realistic failure modes through dynamic adversarial simulation.
• Probing retrieval, memory, tool use, and multi-turn behavior under pressure to surface failure modes that do not appear in static evaluation.

MBZUAI · Supervisor: Dr. Mohammad Yaqub

• Mentored 6 UGRIP interns and actively participated in the BraTS MICCAI challenge.
• Enhanced the MedNeXt base model for BraTS 2023 by integrating a novel schedule-free optimizer introduced by Meta AI in the paper “The Road Less Scheduled”.
• Implemented both greedy and uniform model souping techniques across 18 trained models with varied hyper-parameters to boost performance.
• Awarded the Best Team Award within the UGRIP program.
• Achieved 3rd place in MICCAI's prestigious BraTS 2024 SSA & Pediatrics Challenge and presented our research at MICCAI 2024.

Etisalat · Supervisor: Ahmed Hamada

• Developed a Retrieval-Augmented Generation (RAG) chatbot system that processes Excel datasets to deliver comprehensive data explanations, interactive QA responses, and actionable analytics insights.
• Handled different meetings' requirements and goals with stakeholders.

IIITD · Supervisor: Dr. Saket Anand

• Developed a Drive-by-Wire system for an E-rickshaw, enhancing autonomous control capabilities.
• Calibrated VLP-16 LiDAR and stereo cameras to improve sensor accuracy and reliability.
• Utilized Commonroad Scenario Designer to create lanelet maps for IIIT Delhi OSM map and Carla open drive maps — documentation.
• Implemented a PID controller for throttle and steering control of the E-rickshaw, ensuring smooth and responsive maneuvers.
• Trained and deployed YOLOP for real-time object detection and perception tasks — video.

Jamia Millia Islamia · Supervisor: Dr. Arshad Noor Siddique

• Worked on Linear Friction Actuation to enhance mechanical system performance.
• Designed an image-processing application to analyze photos of friction stir-processed material samples, extracting valuable data for research purposes.
• Automated linear rails using a ball screw, stepper motor, Arduino Uno, and TB6600 motor driver for precise control of a welding torch.