Yash Jha

Intern

Ranchi, Jharkhand, India0 mo experience

AI EnabledAI ML Practitioner

Key Highlights

Led development of a B2B e-commerce platform.
Adapted advanced ML pipelines for healthcare.
Strong foundation in full-stack development and AI.

Stackforce AI infers this person is a Full-Stack Developer with a focus on AI/ML applications in E-commerce and Healthcare.

Contact

Skills

Core Skills

Full-stack DevelopmentProject ManagementMachine LearningResearch

Other Skills

AnacondaArtificial Intelligence (AI)C (Programming Language)C++Cascading Style Sheets (CSS)CommunicationDatabase Management System (DBMS)Deep LearningExpress.jsFew Shot LearningGitGitHubHTML5Information Technology InfrastructureJavaScript

About

I am a final-year B.Tech student at BIT Mesra specializing in Artificial Intelligence and Machine Learning, passionate about building impactful software products and solving real-world problems with code. My experience ranges from developing full-stack web applications (MERN, Next.js) to implementing scalable, production-ready ML pipelines and research workflows. As an SDE intern, I've architected end-to-end platforms, built secure and user-friendly interfaces, and collaborated cross-functionally to ship features that drive measurable impact. My work in academic research and industry has honed my skills in deep learning, optimization, rapid prototyping, and technical problem-solving. I thrive in fast-paced, team-driven environments and am actively seeking challenging software engineering, AI/ML, or product development opportunities where I can learn, contribute, and help deliver solutions at scale.Always eager for new opportunities

Experience

Diptech technologies pvt. ltd.

Intern

Jun 2025 – Present · 9 mos · Patna, Bihar, India · Hybrid

Led the end-to-end development of a full-scale B2B e-commerce platform for large industrial machinery using the MERN stack, from requirement gathering to production-ready architecture.
Translated business requirements into scalable product features, collaborating with stakeholders to define user flows, dashboards, and approval workflows.
Designed and implemented role-based access control (RBAC) with dedicated dashboards for users, admins, and brands, improving operational efficiency and platform governance.
Built a machine-parts visualization and browsing system enabling users to explore components digitally without physical disassembly, enhancing buyer decision-making and user experience.
Developed a technician appointment booking system to strengthen after-sales service, customer engagement, and retention strategy.
Implemented brand onboarding and product listing workflows with admin approval, homepage visibility controls, and business rule validations.
Integrated Razorpay partial payment system with booking-based conditional logic, delivery fee separation, and secure transaction handling.
Automated PDF loan quotation generation to support machinery financing through banking partners, streamlining the sales-to-finance handoff.
Actively contributed to product roadmap discussions, feature prioritization, and release planning based on business impact and user needs.
Coordinated with cross-functional teams to manage task execution, timelines, testing cycles, and deployment readiness gaining hands-on exposure to project management practices.
Architected a scalable backend infrastructure with forward planning for AWS deployment, external UI integrations, and multi-layered security controls.

MERNJavaScriptNode.jsExpress.jsMongoDBProject Management+4

National institute of technology , patna

Research Intern

May 2025 – Jul 2025 · 2 mos · Patna, Bihar, India · On-site

Adapted the IbM2 (CVPR 2024) practical few-shot learning pipeline to medical imaging datasets HAM10K and BreakHis, extending support beyond the original CUB benchmark.
Developed custom utilities to generate and manage few-shot splits (1–16 shots) across multiple datasets, ensuring reproducible experiments.
Replaced naive epsilon search with Bayesian optimization, leading to faster and more stable convergence of the IbM2 training pipeline.
Enhanced training dynamics using cosine-annealing learning rate schedules combined with noise-aware acquisition strategies for robust optimization.
Designed a dynamic classifier head that switches between non-linear (with batch normalization) for higher-shot regimes and linear classification for extremely low-shot settings.
Integrated FEAT-style feature adaptation to boost representation quality and accuracy in low-shot recognition scenarios.
Increased accuracy and stability in higher-shot configurations by fine-tuning classifier depth and dropout regularization.
Refactored core training scripts (search, fine-tuning, classifier) into a modular framework, simplifying experimentation and future extensions.