AI & Backend enthusiast · Student at DTU
Building intelligent
software, end to end.
Student at Delhi Technological University (DTU), building LLM agents, RAG systems and distributed platforms that turn research into working software.
About
I'm a B.Tech student with a major in Electronics and a minor in Computer Science. My focus sits at the intersection of applied AI and backend engineering — large language models, retrieval-augmented generation, computer vision and the systems that make them reliable in production.
- Education
- B.Tech @ DTU
- Aug 2024 — Jul 2028 · CGPA 8.72
Projects
AutoExamAI
2026An end-to-end LLM and RAG-powered assessment platform that automates exam generation and student evaluation.
- AI-powered assessment engine that turns course syllabi into complete exams with configurable MCQ, short- and long-answer distributions, cutting prep time by 90%+.
- Automated grading framework using RAG, semantic similarity scoring and rubric-guided evaluation, achieving 90%+ agreement with instructor grades.
- Evaluates 100+ submissions in minutes against generated answer keys.
Lumen GridFusion
2026An open-source framework bridging LLM agents with multidimensional scientific datasets via SQL-queryable interfaces.
- Exposes Xarray-backed data through SQL-queryable interfaces, enabling natural-language exploration of climate, geospatial and Earth-system data.
- AI-powered query recommendation and Apache DataFusion execution pipelines for NetCDF and Zarr datasets.
- Achieved a 34× improvement in large-scale geospatial analytics via optimized SQL-level spatial aggregation.
TraceForge
2026An observability and self-healing platform for distributed microservices with automated monitoring and recovery.
- 6-service microservice architecture with distributed tracing, structured logging and dependency-aware failure propagation for production failure simulation.
- Automated monitoring and recovery pipeline using Prometheus metrics, reducing Mean Time To Resolution (MTTR) by over 70%.
- Near real-time recovery from service failures across the stack.
Experience
- To resolve the issue of low-light, skewed smartphone uploads of student test papers, engineered an end-to-end Document AI pipeline implementing Otsu’s Binarization and OpenCV perspective transforms for systematic document deskewing and normalization.
- Faced with unstructured exam sheets that lacked clear boundaries for automated grading, architected a Document Layout Analysis (DLA) framework using a fine-tuned YOLOv11-nano model to segment pages into precise bounding boxes isolating student answers, question numbers, and teacher marks.
- Overcame the high error rates caused by handwriting variance and cursive text degradation in baseline OCR by deploying a Vision Transformer (TrOCR) via Hugging Face for highly accurate handwritten text extraction.
- Technologies Used: Python, OpenCV, PyTorch, Hugging Face (TrOCR), YOLOv11, FastAPI, Pydantic, Instructor, Sentence-Transformers, Docker, Ollama.
Tools & Framework
Languages
- Python
- JavaScript
- TypeScript
- SQL
- C++
AI / ML
- TensorFlow
- LangChain
- LangGraph
- HuggingFace
- Pandas
- NumPy
Backend
- Node.js
- Django
- FastAPI
- Socket.IO
- PHP
Frontend
- React
- Next.js
- Angular
Databases
- MongoDB
- PostgreSQL
- Redis
- Vector DB
DevOps / Cloud
- Kubernetes
- Docker
- AWS
- GCP
Contact
Let's build something thoughtful together.
Open to internships, research collaborations and interesting problems in AI and backend engineering.