AI ENGINEER · BANGALORE, INDIA
Richard Samuel
Building agentic AI systems at MeshDefend.
Published research in remote-sensing restoration
and adversarial robustness.
Publications
Remote sensing restoration and adversarial ML robustness, grounded in an ISRO/NRSC research internship.
-
PUBLISHED
Enhanced Multi-Scale Pyramid Deep Image Prior for Unsupervised Remote Sensing Image Restoration
Unsupervised restoration of satellite imagery using a multi-scale deep image prior, removing the need for paired clean/corrupted training data. Developed during a research internship at ISRO/NRSC.
IEEE Xplore ↗ View code ↗ -
PRESENTED
FedHyperGNN: Temporal Hypergraph Neural Networks for Privacy-Preserving Federated Recommendation Systems
A federated learning approach using hypergraph neural networks to model higher-order user-item relationships in recommendation systems, without centralizing user data.
-
BEST PAPER AWARD
Reinforcement Learning Based Framework for Dynamic Defense in Adversarial Environments
A PPO-based reinforcement learning framework for defending medical imaging classifiers against adversarial perturbations.
View code ↗ -
PRESENTED
CAT-SR: Cross-Market Attention Transfer for Seller Recommendations
Attention-based transfer of seller-recommendation signal across markets.
Research, interactable
A conceptual, in-browser visualization of my GRSL work: unsupervised restoration of corrupted satellite imagery. Procedurally generated scene, real PSNR math, zero servers.
PSNR —
Selected Projects
Applied AI services and full-stack ML, shipped end to end.
aRAG
Serverless retrieval-augmented generation platform: upload documents and converse with them through AWS Lambda functions, Cognito authentication, Pinecone vector search, and Redis-cached DynamoDB storage behind a React interface. This site's chat widget runs on it.
FinAL
AI stock analysis: LSTM 7-day price forecasting, BERT sentiment on news headlines, and a Gemini-powered expense chat, over live market data from Finnhub and yfinance.
PixelPerfect
AI image studio: generate images from text prompts, caption them with ViT-GPT2, summarize them with Gemini vision, and sharpen photos. FastAPI backend.
SureScan
Brain tumor detection and diagnosis: YOLOv11 localization, EfficientNet classification across 44 classes, XGBoost survival prediction, and Gemini-powered report Q&A.
Demos are hosted on free tiers and may take ~30s to wake on first load.
About
AI Engineer at MeshDefend, an AI-powered IT services company for data infrastructure. Currently full-time, following a six-month internship there from December to May. Shipped six automation workflows to production and contributed to a system-design feature with a provisional patent filed.
B.E. in Computer Science (AI & ML specialization) from PSG College of Technology. Research in Machine Learning and Deep Learning including satellite imagery restoration using a Deep Learning model during an ISRO/NRSC internship, and adversarial robustness work using reinforcement learning on sensitive images.
At PSG Tech, served as Events & Sponsorship Head (2023–24) and Technical Team Member (2024–25) of The Eye, a student club. Student Member of the IEEE Geoscience and Remote Sensing Society. Received the Technical Excellence Award from the Computer Science Engineering Association (2026), and placed 3rd at Hacksphere, 2nd and 3rd at Ideathon, with a special mention at the Caterpillar Codeathon.
Beyond work
Interests outside research and engineering.
- Photography Selected photographs on VSCO ↗
-
Games Console gaming, with an interest in interactive system design.
-
Robotics Embodied agents: a long-term research interest.
Get in touch
Open to conversations on agentic AI systems, applied ML, and research collaboration.
Start a conversation