// Research · Architecture · Scale

Dr. Amarnath R.

Head of AI Engineering  |  Consultant Organization Tech Optimization  |  Principal AI Architect  |  Ph.D. in Computer Vision  |  GenAI & Multi-Agent Systems

Bridging deep academic research with enterprise-scale engineering. I architect Agentic AI systems and Deep Learning pipelines that move beyond POCs into production-grade, high-ROI innovation.

15+
Years Experience
$3M+
Funding Secured
50+
Global Clients
21+
Publications

At a Glance

💰

Funding & Leadership

Principal Investigator for a COVID-19 thermal screening AI project, securing $3M+ in DST (Govt. of India) and VC funding.

⚙️

Enterprise Scale

Architected LLM-powered automation platforms for 50+ global clients (including Bosch), delivering 40% operational efficiency gains.

🚀

Entrepreneurship

Founder of Codeidea.in — competitive coding & hackathon platform with 500+ students. Acquired by IVISLabs Pvt. Ltd.

🧠

Core Tech Stack

LangGraph, LangSmith, OpenClaw, PyTorch, Transformers, Kubernetes, AWS.

Academic & Applied Research

My Ph.D. thesis, Segmentation in Compressed Document Images (University of Mysore, 2019), forms the foundation of my applied work in visual AI and OCR pipelines.

arXiv 2026 · Preprint

HMS-VesselNet: Hierarchical Multi-Scale Attention Network with Topology-Preserving Loss for Retinal Vessel Segmentation

A hierarchical multi-scale network processing fundus images across four parallel branches at different resolutions, combining Dice, BCE, and centerline Dice losses with hard example mining. Achieves mean Dice of 88.72%, Sensitivity 90.78%, and AUC 98.25% across DRIVE, STARE, and CHASE_DB1.

arXiv:2603.21891 [eess.IV] · 19 pages, 14 figures, 8 tables
Under Review

Input Resolution in Retinal Vessel Segmentation

An empirical study across five datasets quantifying the structural loss of microvascular details during downsampling.

Under Review: JIIM
IEEE Access · 2024

Hard Exudates Segmentation in Diabetic Retinopathy Using DiaRetDB1

Published in IEEE Access, vol. 12, 2024. DOI: 10.1109/ACCESS.2024.3455433

IEEE Access
SCIE · SCOPUS · DBLP

Automatic Localization and Extraction of Tables from Handheld Mobile-Camera Captured Handwritten Document Images

Published in Journal of Intelligent & Fuzzy Systems (JIFS), vol. 36, no. 3, pp. 2527-2544, 2019.

Journal of Intelligent & Fuzzy Systems
Applied Research

Automated Content Extraction

Developed advanced OCR and NLP models reducing manual cataloguing effort by 70%+ for e-commerce pipelines at Amazon, Flipkart, and LiveAuctioneer.

  • 2026 HMS-VesselNet: Hierarchical Multi-Scale Attention Network with Topology-Preserving Loss for Retinal Vessel Segmentation. Amarnath R. arXiv:2603.21891.
  • 2024 Hard Exudates Segmentation in Diabetic Retinopathy Using DiaRetDB1. Ma Yinghua, Yang Heng, R. Amarnath, Zeng Hui. IEEE Access, vol. 12.
  • 2023 Pruning distorted images in MNIST handwritten digits. Amarnath R., Vinay Kumar V. CoRR, abs/2307.14343.
  • 2022 Segmentation of pectoral muscle in mammograms using granular computing. B. V. Divyashree, Amarnath R., et al. J. of Information Technology Research, vol. 15, no. 1.
  • 2021 Detection and classification of hard exudates in retinal images. Thamer Al Sariera, Lalitha Rangarajan, Amarnath R. J. of Intelligent & Fuzzy Systems, vol. 38, no. 2.
  • 2021 An experimental study on the effect of noise in CCITT Group 4 compressed document images. A. N. Sukumara et al. Advances in AI and Data Engineering.
  • 2019 Regression-based sub-image matching methodology for recognizing an Indian paper bill. Amarnath R., P. Nagabhushan. IJEEA.
  • 2019 Word and character segmentation directly in run-length compressed handwritten document images. Amarnath R., P. Nagabhushan, Mohammed Javed. CoRR.
  • 2019 Automatic localization and extraction of tables from handheld mobile-camera captured handwritten document images. Amarnath R. et al. JIFS, vol. 36, no. 3. (SCIE, SCOPUS, DBLP)
  • 2018 Text line segmentation in compressed representation of handwritten document using tunneling algorithm. Amarnath R., P. Nagabhushan. IJISAE, vol. 6, no. 4.
  • 2018 Novel approach to locate region of interest in mammograms for breast cancer. B. V. Divyashree, Amarnath R. et al. IJISAE, vol. 6, no. 3.
  • 2018 Enabling text-line segmentation in run-length encoded handwritten document image using entropy-driven incremental learning. CVIP-2018, IIIT Jabalpur. (IAPR Endorsed)
  • 2018 Entropy-based approach for enabling text line segmentation in handwritten documents. Springer LNNS, DAL'18, Mysore.
  • 2018 Line detection in run-length encoded document images using monotonically increasing graph model. IEEE ICACCI, Bangalore.
  • 2017 Segmentation of handwritten text document directly in compressed image version. SSDA, Jaipur. (IAPR Endorsed)
  • 2017 Spotting separator points at line terminals in compressed document images for text-line segmentation. IJCA, vol. 172, no. 4.
  • 2009 Fingerprint Biometrics with movement enhanced technique. NCCCIS-2009, Coimbatore.
  • 2009 Image Processing over IP Networks. NCCCIS-2009, Coimbatore.
  • 2008 Magic at Zero desk-space: Designing and implementing an advanced human-computer interaction system. MNGSA, Karunya University.
  • 2008 An optimized string searching algorithmic technique. MNGSA, Karunya University.
  • 2008 Designing a novel web-based e-learning authoring tool for dynamic content management. MNGSA, Karunya University.

Career Timeline

15+ years defining AI strategy, system architecture, and product roadmaps for enterprise and government clients.

Founder & Developer

Codeidea.in (Acquired by IVISLabs Pvt. Ltd.)
Jun 2024 – Jul 2025 · Sabbatical
  • Built competitive coding and problem-solving platform for students and developer communities
  • Algorithm visualizations and real-time problem-solving environments
  • Organized 3+ hackathons engaging 500+ students across multiple universities

Solutions Architect

HCL Technologies Ltd.
Jul 2022 – May 2024 · Chennai
  • Architected AI-powered enterprise applications for 50+ global clients, integrating LLMs and GenAI
  • Partnered with Bosch on LLM-powered automation — intelligent document processing and knowledge retrieval
  • 40% operational efficiency improvement; 80% customer satisfaction increase via real-time AI analytics
  • Led cross-functional teams of 8+ engineers deploying production-ready AI systems

Academic Foundation

Doctorate

Ph.D., Computer Science

University of Mysore · 2016 – 2019
Thesis: Segmentation in Compressed Document Images. Novel research on document image segmentation, layout analysis, and recognition systems.
Master's

M.Tech., Computer Cognition & Technology

University of Mysore · 2011 – 2013
Thesis: Indian Paper Currency Recognition Systems for Visually Impaired.

Technical Toolkit

Languages

Python Java JavaScript MATLAB

AI / ML Frameworks

PyTorch TensorFlow Transformers Scikit-learn

Agentic AI / LLM

LangGraph LangChain LangSmith OpenClaw

Full Stack

FastAPI Flask ReactJS REST GraphQL

Cloud & DevOps

AWS Docker Kubernetes CI/CD

Databases

PostgreSQL MongoDB Pinecone ChromaDB

Hammering Info for Dummies

Understanding RAG Systems the Simple Way — A guide for those who prefer clarity over jargon.

What Is RAG? (Really, Really Simple)

Think of RAG like asking someone to look something up in a book before answering your question, instead of relying only on their memory.

RAG = Retrieval-Augmented Generation. In plain English: the AI reads your documents first, then answers your questions based on what's written there.

"You ask your doctor a question about your test results. Instead of going from memory, they pull out your file, read it carefully, and then answer based on the facts. That's RAG."

Why Should You Care?

  • Accuracy: The AI doesn't guess. It reads your actual documents.
  • Privacy: Your documents stay on your computer. No cloud, no sending files elsewhere.
  • Trust: You can see where the answer came from — which page, which file.
  • No subscriptions: Once set up, it works without monthly fees.

Especially useful if you're managing family documents, medical records, financial papers, or business files.

How Does It Actually Work? (Three Steps)

  • Preparation: You give the system your documents (PDFs, Word docs, etc.). It reads them once, converts the text into a special format, and saves everything locally.
    Like organizing your filing cabinet for the first time — takes effort upfront, but then everything is organized.
  • You Ask a Question: You ask the system something like "What are my treatment options?" or "What does my policy say about refunds?"
  • The System Answers: It finds the most relevant parts of your documents and gives you an answer with the source. You can verify it against the original document.
    Like a librarian who finds the exact book, page, and paragraph that answers your question.
  • Real-Life Examples

    Medical: Upload all your medical reports and ask "Has my cholesterol improved?" The system reads all your lab reports and gives you facts.

    Financial: Upload your insurance documents and ask "What happens if I can't pay?" Find the exact clause in your policy.

    Family Records: Upload grandchild's school deeds and ask "What are the academic requirements?" Get instant answers without calling.

    The Biggest Benefit?

    You're no longer at the mercy of your memory. You have a system that reliably reads, understands, and retrieves information from your documents whenever you need it.

    It's like having a patient librarian in your home who never gets tired, never forgets, and always tells you the source.

    Learn More →

    RAG for Dummies in Action

    See how Retrieval-Augmented Generation works in a real, live demonstration. Watch the system retrieve information from documents and generate answers.

    In this demo, you'll see the complete RAG workflow: uploading a document, asking specific questions, and getting answers grounded in the actual content with source attribution.

    🚀 Try it yourself: No installation required. Visit our live interactive demo on HuggingFace Spaces and practice with your own documents instantly.

    Try Live Demo on HuggingFace Spaces →

    Live Demos & Projects

    Interactive proof-of-concepts showcasing applied AI capabilities and engineering depth.