Greater Delhi Area
Passionate Computer Science Engineering student with a strong interest in Data Structures & Algorithms, Operating Systems, Web Development, Machine Learning, NLP, and Database Management Systems. Always eager to expand my technical expertise and enhance my creativity in these domains. ✔ Interned at Google Cloud in summer of 2025 ✔ Worked in 3 research labs during B.Tech degree ✔ Solved 600+ DSA problems on LeetCode and Codeforces ✔ Codeforces Rating: 1357 (Pupil) Driven by a love for problem-solving and innovation, I am constantly exploring new technologies to refine my skills and contribute meaningfully to the tech community. 🚀
Worked on optimization of tail latency for heterogeneous edge devices (Guide: Dr. Arani Bhattacharya) - SafeTail 2.0 is an intelligent workload scheduling framework for heterogeneous edge computing environments that uses Reinforcement Learning (RL) to dynamically allocate service requests across edge servers. - This project extends the SafeTail 1.0 framework by introducing a Deep Reinforcement Learning–based controller that learns optimal scheduling strategies based on system state, workload characteristics, and queue delays. - Built a DQN-based reinforcement learning agent for tail latency optimization on heterogeneous edge devices, including reward design (step- and episodic-level), data preprocessing, state construction, and model/encoder design for variable-length system states. - Designed and implemented a controller pipeline for scheduling requests between users and heterogeneous edge servers, incorporating request queuing/dequeuing, dispatching, and coordination to enable episodic RL training.
- Developed a GCA (Gemini cloud assist) chat feature within Google Cloud Console to diagnose database performance issues and provide insights. - Developed an engine agnostic pipeline along with trend and correlation analysis to answer complex queries. - Integrated with Gemini 2.5 to provide fast responses.
Developed a recipe generation model similar to RecipeGPT.io, leveraging the Recipe1M and RecipeNLG datasets. Our approach involves: • Data reduction using KMeans clustering (SpaCy embeddings) and Faiss-based clustering (DistilBERT embeddings), reducing dataset from 1 million to 100K recipes. • Applied a pre-made Named Entity Recognition (NER) model to extract structured info from dataset. • Fine tuned GPT-2 on structured dataset to generate structured recipes
Rema is a professor-student interaction app designed to streamline communication between student researchers and professors. It enables professors to manage team members (add, view, delete), share research materials, and post project meeting notes via real-time messaging using WebSockets. The app integrates Google OAuth for authentication and incorporates the Gmail API to allow professors to access their starred emails within the platform.
I conducted research under Dr. Dhruv Kumar on prompt engineering, evaluating the performance of GPT-3, Copilot, and Gemini on undergraduate computer science assignments. Our study focused on analyzing their effectiveness in generating solutions, identifying limitations, and exploring strategies to optimize responses. The research aligns with the findings presented in this paper, which examines the capabilities of large language models in solving academic tasks.