Sep 2024 – Jan 2025 · 4 mos · Delhi, India · On-site

• As an AI/ML Intern at IIT Delhi, I worked on a World Food Programme (WFP) project focused on building a cost-efficient, reliable, and intelligent surveillance system for food security monitoring across multiple warehouses in India.
• I contributed to designing a Cost-Optimized Surveillance System by leveraging Jetson Edge Computing and Dockerized microservices, reducing deployment costs by nearly 60% across 8 warehouses. I implemented real-time motion detection using ResNet models on the Jetson Orion Nano device, achieving 97% accuracy with ~800ms latency, which further triggered YOLOv8-based anomaly detection for activity monitoring.
• To ensure system robustness, I engineered an Edge-to-Cloud Reliability Architecture, deploying Dockerized camera services with automated reattachment (3-retry logic) integrated with Firebase, successfully managing 37 concurrent image streams across 8+ edge devices.
• This experience gave me deep exposure to computer vision, edge AI deployment, microservice architectures, and cloud-edge integration, while working on a project that directly contributed to global food security goals.

Jun 2024 – Sep 2024 · 3 mos · India · Remote

• As an NLP Research Intern at DBlockchainers, I worked on building an end-to-end NLP experimentation framework to train, evaluate, and optimize a wide range of language models. Over the course of my internship, I implemented and trained models from the ground up, starting with classical embeddings like Word2Vec, GloVe, and FastText, and progressing to transformer-based architectures such as BERT and RoBERTa, all within a PyTorch Lightning framework.
• A major focus of my work was on fact-checking datasets such as ChartFact and ChartCheck, where I experimented with embedding fusion techniques (e.g., combining GloVe and FastText) to improve the representation of tabular and textual data. I also developed a large-scale evaluation engine capable of testing models across multiple datasets, using Hydra for configuration management and strategies like negative sampling to enhance model generalization and recall.
• On the deployment side, I optimized trained models for real-world inference by applying ONNX Runtime-based quantization and other efficiency techniques, significantly reducing latency while keeping accuracy stable. I also incorporated preprocessing and tokenization pipelines using NLTK and Hugging Face Transformers, ensuring that the models could handle diverse input formats consistently.
• This internship deepened my expertise in transformer architectures, NLP evaluation pipelines, and model optimization for production, while allowing me to explore the research side of fact verification and information consistency in natural language processing.

Jun 2024 – Jul 2024 · 1 mo · Online · Remote

• As part of the IBM Front-End Web Development (FEWD) SkillBuild Program, I collaborated with a team on “Water Watch”, a platform that enables local communities to raise and track water-related issues. I specifically worked on building the issue management pages in React, where users could upload images, raise tickets, and check issue status. This project not only strengthened my front-end development skills but also gave me valuable exposure to team collaboration, problem-solving, and building user-focused solutions.

Mar 2024 – May 2024 · 2 mos · Delhi, India · Remote

• As an AI/ML Engineer at Xander, I worked on building a no-code machine learning platform where users could simply upload datasets, and the system would automatically identify the best-suited model and generate it for them. My primary contributions included automating regression and classification tasks, streamlining the model selection and training pipeline, and integrating conversational interfaces.
• I also developed and integrated a chatbot using Botpress, enabling users to interact with the platform more intuitively and receive real-time guidance throughout the ML workflow. This experience strengthened my skills in end-to-end ML automation, chatbot development, and no-code platform design, while contributing to making AI accessible for non-technical users.