Scroll to explore
I’m a Computer Science undergraduate at IIIT Allahabad with a strong focus on machine learning research and Agentic systems. My work centers around building deep learning models that are not only accurate, but interpretable and deployable in real-world settings.
I’m particularly interested in neural networks, multilingual AI safety, and model optimization. My approach combines strong theoretical foundations with hands-on experimentation,from implementing research papers to engineering full-stack ML systems.
Beyond coursework, I actively work on research-driven projects, participate in hackathons, and build end-to-end AI applications. I’m motivated by problems that sit at the intersection of research rigor and practical impact.
B.Tech in Information Technology
IIIT Allahabad(2024-2028)
Currently working on multilingual safety probes using whiteboxing techniques, scalable ML systems for real-world deployment, and optimization techniques for efficient deep learning models.
Research Focus
Interpretable & Multilingual AI
Specialization
Deep Learning Systems & AI Safety
Co-authored and presented a workshop paper at AAAI-26 (AICS Workshop) on cross-lingual transfer and multilingual AI safety.
Maintained a CGPA of 9.3 at IIIT Allahabad while actively contributing to research and ML system development.
Top finalist in national-level AI hackathon(Educathon 2.0), building full-stack ML solutions under tight constraints.
Maintained and managed a project repository during OpenCode (national-level open source competition), overseeing contributions, reviewing updates, and guiding feature development.
Core member of GeekHaven (AI/ML Club, IIIT Allahabad), mentoring students, conducting workshops, and leading discussions on deep learning and applied Agentic systems.
Collaborated with professionals from Google Germany and ETH Zurich on research-driven AI initiatives and AI safety exploration.
Research & Publications
AAAI-26 AICS Workshop, Singapore
We propose a cross-lingual transfer pipeline that adapts safety-aligned behaviors from high-resource languages to low-resource settings using bilingual contrastive objectives and consistency constraints. The approach improves robustness on multilingual safety benchmarks while preserving task performance, enabling safer deployment across diverse language environments.

Built a real-time air-writing system using MediaPipe for fingertip tracking and CNN-based symbol recognition. Implemented parsing logic for multi-digit expressions and exploring CRNN + symbolic evaluation integration.
Developed a backend system to parse bank emails (IMAP) and extract transaction details using fuzzy matching and early LLM based financial insights. Built secure dashboard with analytics and categorization pipeline.
CNN-based OCR system for noisy CAPTCHA text. Implemented heavy augmentation (rotation, distortion, noise) and optimized Flask backend for sub-200ms inference with live React UI integration.
Built an agentic LLM pipeline using LangGraph and FastAPI to generate interpretable stock insights. Integrated FinBERT sentiment analysis, sector benchmarks, and structured scoring for profitability, growth, valuation, and cash flow.
A curated visual walkthrough.
1 / 5
AirWriting demo: real-time math expression input and solver.