I’m a CS undergrad at Manipal Academy of Higher Education and I spend most of my time building things that work at scale. I’ve built distributed message brokers that handle 100K+ messages per second, VLM evaluation pipelines with systematic prompt engineering, multimodal deep learning systems trained on millions of data points, and real-time computer vision systems deployed in corporate environments.

My Path into Engineering

I got into tech because I genuinely wanted to know how things work under the hood. Not just using tools, but actually understanding them well enough to rebuild them from scratch. That curiosity is what pushed me to dig into consensus algorithms, build my own message brokers, design VLM evaluation frameworks, and tear apart neural network architectures to see what makes them tick.

At Manipal, I’ve kept an 8.7 CGPA while focusing heavily on building real projects. Not coursework projects, but actual production systems with proper architecture, tests, and CI/CD pipelines.

Building Real Systems

I think the best way to actually learn distributed systems is to build one yourself. So I wrote a Kafka-inspired message broker from scratch in Go with Raft consensus, gRPC transport, log replication, and segment-based storage. No external libraries for the core. Just me reading the papers and implementing them.

On the AI/ML side, I built a Driving Scene Description Generator that processes autonomous driving images through Vision-Language Models with 8 systematically designed prompt variants and evaluates the outputs against ground truth using 8 different metrics. The most valuable part was building the AI error analysis agent that detects systematic failure patterns and auto-generates improved prompts.

I also built PriceScope, a multimodal deep learning system trained on 1.48 million Mercari listings. It fuses BiLSTM text encoders with categorical embeddings for price prediction, complete with Optuna tuning, SHAP explainability, and ONNX export. The full stack runs with FastAPI, Next.js, and MongoDB.

And NeuralRAG, a self-correcting RAG pipeline that validates its own outputs, catches hallucinations, and automatically reformulates queries when the answers aren’t good enough. I learned more from debugging the failure modes than from getting the happy path working.

Industry Experience

During my internship at Reliance Industries Limited in the Video Analytics Division, I built real-time occupancy detection systems using YOLOv11 and DeepSort, processing 10+ concurrent RTSP camera feeds at 60+ FPS. The most interesting part was figuring out multi-camera spatial deduplication to avoid double-counting people visible from overlapping camera angles.

What I Care About

I’m most interested in problems where systems engineering meets machine learning. The kind of work where you need both algorithmic depth and solid infrastructure thinking. Whether it’s making a message broker fault-tolerant, designing VLM evaluation pipelines, or building production ML systems end-to-end, I like the challenge of making things work reliably when it matters.

Currently exploring: VLM prompt engineering, production ML pipelines, and distributed systems at scale.

Small Achievements and Certs

• 300+ problems solved on LeetCode & CodeForces
• Top 10 at Honeywell SDG Hackspace 2025
• Deep Learning Specialization (Coursera)
• Microsoft Azure AI Fundamentals certified