Shubham Prakash

I’m a final-year ECE student with 3 production internships spanning Computer Vision, Generative AI, and end-to-end ML deployment. I build production-ready systems that turn model quality into reliable APIs using real observability, load testing, and CI/CD.

In my Generative AI Intern role at Bridged.Media, I improved LLM response quality by 20–30% using AI agents and advanced prompt engineering in customer-facing pipelines. I shipped 5+ production APIs via GitHub Actions CI/CD and diagnosed/resolved 8+ critical bugs in email automation, while authoring Postman collections covering 100+ endpoints.

In Computer Vision at KalkiNi, I architected a YOLO-based real-time object detection and billing system achieving 95%+ detection accuracy—reducing billing time by 80% and eliminating manual checkout steps. I also built an automated YOLO annotation and data pipeline that cut labeling time by 85%, and worked on model optimization, IoU tuning, confidence threshold calibration, and low-latency inference.

As an AI Developer Intern at Sarvm.ai, I fine-tuned speech AI models on 22,500 synthetic audio samples to increase accuracy by 12% and reduce error rates by 10%. I developed a multilingual OCR pipeline (PaddleOCR + Tesseract) for 13 languages with 99%+ extraction accuracy and created an LLM-powered structured JSON parser reaching 90% accuracy on noisy documents.