AI-driven applications. Resilient automations.
I build the steps between an idea and what comes next.
AI / ML diploma candidate · seeking a technology internship
the tools I work with, and the things I build myself
I’m Kushal M Anvekar, a Computer Science (AI/ML) diploma candidate at M.S. Ramaiah Polytechnic. I work across the development lifecycle—from model training to containerized services and cloud deployment.
AI web applications, deep learning inference, and local RAG systems. Built to turn technical ideas into something useful.
PyTorch · LLMs · Computer visionFrom a clean frontend to a resilient backend. Containers, cloud environments, and automated workflows connect the pieces.
Rust · Docker · GCP · n8nSeeking a technology internship in scalable software development and secure application practices. Bengaluru, India.
Read my resumeA connected stack, from data to deployment.
Every layer has a job. Every handoff matters.
Train models, retrieve the right context, and connect inference to a real use case.
Build the interface, containerize the backend, and automate the steps that should run themselves.
Languages, frameworks, data, and infrastructure. A toolkit for building across the entire system.
Explore the repositoriesApplied AI, useful tools, and the next experiment.
The work—and the code behind it.
Security tooling for developers and AI coding agents. Audit code offline, trace input flows across functions, and verify patches against Git changes or saved baselines. Explore findings in a terminal workspace or connect agents through MCP. Optional explanations use Ollama, NVIDIA NIM, or both.
Building a network where AI agents collaborate on tasks, receive payments, and build verifiable reputation on Monad. The project brings task coordination, on-chain settlement, and a visible history of completed work together.
Developing a portable, model-agnostic runtime for domain-specific AI systems. Bedrock provides execution, tools, memory, policies, and workflows while letting each application define its own identity and behavior.
A research workspace for NSE equities with stock exploration, watchlists, market context, and sourced research assistance. Data freshness is made explicit throughout the experience. The portfolio demo uses illustrative data and a rule-based assistant.
Planning a browser-based toolkit for images, PDFs, archives, and developer tasks. The focus is a consistent workflow for processing files locally, inspecting outputs, and passing results between compatible tools.
3D spatial understanding from a single image. Researching how to locate objects at real-world scale and understand which objects rest on or support others. The aim is a more useful picture of a scene’s physical layout.
one image → a spatial sceneMUTE — Mask-free Unified Token Editing. Investigating precise, localized edits to AI-generated images, such as changing the color of an object while preserving the surrounding scene. The goal is finer control over what changes and what stays intact.
one focused edit → the same surrounding sceneHave an opportunity, an interesting problem,
or something you’d like to build? Let’s talk.