Jaipur, India
Hi, I'm Kartik Mantri
AI/ML Engineer & Decentralized Systems Builder
CS undergrad working across applied AI/ML and Web3 — from segmentation models built from scratch in PyTorch to smart contracts settling on-chain micro-payments.
About
I'm a Computer Science undergraduate at LNMIIT, working across two things that don't usually share a portfolio: applied AI/ML and decentralized systems.
On one side, that means implementing deep-learning architectures from scratch — U-Net variants, Kolmogorov–Arnold Networks — and benchmarking them on real medical imaging data. On the other, it means designing smart contracts and on-chain payment flows for Web3 products.
I like the full stack of a hard problem: reading the paper, implementing the layer by hand, and shipping the product it powers.
Education
The LNM Institute of Information Technology (LNMIIT)
Bachelor of Technology in Computer Science and Engineering
Jaipur, India
Experience
Research Intern — Deep Learning & Medical Image Segmentation
PDPM IIIT Design & Manufacturing (IIITDM) Jabalpur
Jabalpur, India
Jun 2026 – Jul 2026
- Built and benchmarked segmentation architectures — U-Net, U-KAN, two-stage W-Net, and dual-decoder Y-Net — for joint optic-disc/cup segmentation on retinal fundus images (glaucoma screening via cup-to-disc ratio).
- Achieved a test mean Dice of 0.9165 (target 0.90) with a dual-decoder U-Net using cross-attention and a custom containment loss.
- Implemented Kolmogorov–Arnold Network (KAN) layers from scratch in PyTorch (B-spline basis via Cox–de Boor recursion) as a tokenized bottleneck, benchmarked across REFUGE, G1020, and ORIGA.
Featured Projects
EarningsEdge
AI Earnings-Call Analyzer
An AI platform that ingests BSE/NSE earnings-call PDFs and extracts analyst-grade signals for retail investors — sentiment vs. last quarter, management-tone dimensions, red-flag language, and significance-ranked key quotes.
- Engineered a dual-LLM pipeline with automatic fallback (Groq Llama-3.3-70B → Gemini 2.0 Flash Lite) for resilient, low-latency inference.
- Shipped the full stack end-to-end — ingestion, analysis, and UI — during the Replit Buildathon.
Glaucoma Segmentation Research
Optic Disc/Cup Segmentation from Scratch
Research codebase benchmarking U-Net, U-KAN, W-Net, and dual-decoder Y-Net architectures for joint optic-disc/cup segmentation on retinal fundus images, used for glaucoma screening via cup-to-disc ratio.
- Dual-decoder U-Net with cross-attention and a custom containment loss reached a 0.9165 test mean Dice.
- Kolmogorov–Arnold Network layers implemented from scratch in PyTorch as a tokenized bottleneck.
PayPerUseAI
Decentralized Pay-Per-Request AI Platform
A decentralized AI service platform integrating on-chain micro-payments (x402 protocol) with model access control — trustless, pay-per-inference API consumption without centralized billing.
- Designed Monad smart contracts for request validation, payment verification, and wallet-based access authorization.
- Implemented the x402 payment mechanism for seamless on-chain micro-payments at scale.
Forge
Full-Stack Productivity & DSA Practice Platform
A full-stack productivity platform combining a React frontend with a FastAPI backend — goals, milestones, tasks, notes, resources, and DSA practice in one unified workspace.
- Modular FastAPI backend with dedicated routers for goals, tasks, notes, and DSA problems, with auto-generated API docs.
- React + TypeScript + Vite + Tailwind frontend for goal tracking, milestone planning, and an interactive DSA practice section.
LNMIIT_LMS
Library Management System
A library user-management system with role-based authentication for managing book inventory, borrowing, and user roles.
- Role-based authentication distinguishing librarian and student access.
- Full JavaScript stack for the library workflow end-to-end.
Technical Skills
Languages
AI / LLMs
Deep Learning
Web & App Development
Blockchain / Web3
Tools & Workflows
Leadership & Involvement
Organizing Committee Member — LNMHacks 7.0 & 8.0
Public Relations Team
Jaipur, India
Nov 2024 – Present
- Led sponsor, participant, and volunteer communication and promotional strategy for LNMIIT's flagship hackathon across both editions.
Let's Talk
Working on something in AI/ML or Web3, or just want to talk shop? My inbox is open.