Agent Systems
Agent-first products where autonomous agents gather, structure, and act on data themselves — Mindstruck, built on Eve, Vercel's filesystem-first runtime for durable agents.
KANISH PAUL
Developer Portfolio
Profile
Kanish Paul builds agentic AI systems — software where models plan, call tools, and act on real state instead of only answering prompts. He is a B.Tech Computer Science and Engineering student specializing in Artificial Intelligence and Machine Learning at the Institute of Engineering and Management, working across agent runtimes, multi-model architectures, and the full-stack surfaces that make agent behaviour inspectable.
His current agent work includes Mindstruck, an agent-first CRM built on Vercel's Eve framework where background agents gather and structure data on their own; Chess-Laya, a hybrid engine running Minimax, Laya, and Gemini 2.5 Flash in parallel with sub-second decisions; and an autonomous search-and-rescue drone that perceives and decides at the edge at 28–30 FPS.
The supporting stack is Java, Python, TypeScript, JavaScript, SQL, Solidity, C, HTML5, CSS3, Next.js, React, Spring Boot, PyTorch, OpenCV, Tailwind CSS, and TensorFlow, with Git, GitHub, AWS, Linux/WSL, Remix, Render, and Railway used to ship and operate the systems his agents touch.
Work
Multi-Model Orchestration
Deterministic engines and generative models decided side by side, as in Chess-Laya, where Minimax, Laya, and Gemini 2.5 Flash run in parallel with real-time state synchronization.
Autonomy at the Edge
Perception and decision-making on-device: a search-and-rescue drone running real-time human and obstacle detection at 28–30 FPS, plus escrow and dispute logic that settles without a trusted middleman.
FAQ
What does Kanish Paul build?
Agentic AI. Autonomous agents, multi-model architectures, and the full-stack interfaces around them — everything here is an agent system, a hybrid decision engine, or an autonomous device.
Which agentic projects are featured here?
Mindstruck, an agent-first CRM on Eve; Chess-Laya, a hybrid Minimax, Laya, and Gemini 2.5 Flash engine; AXI, with autonomous escrow and dispute resolution; and an autonomous search-and-rescue drone. The Career page holds the complete résumé.
Why the focus on agentic AI?
Agents are the part of AI that actually does work — gathering context, deciding, then acting. Building them needs the whole stack: model orchestration, durable state, tool access, and a concise interface a person can trust and correct.