Applied AI Engineer | Geospatial & Remote Sensing | ML Systems | LLMs | Edge AI
📍 Singapore
🎓 M.Tech in Intelligent Systems — National University of Singapore (NUS)
🛰️ Former Scientist/Engineer — ISRO (Indian Space Research Organisation)
🔗 LinkedIn: https://linkedin.com/in/yatharthsant
I’m an Applied AI Engineer with 6+ years of experience building end-to-end machine learning systems—from research prototypes to production-grade deployments.
My work sits at the intersection of AI, geospatial data, and performance-critical systems:
- Deep learning for remote sensing & SAR imagery
- Raster / satellite data pipelines at scale
- LLMs & RAG systems for real-world automation
- Edge AI & hardware-aware deployment (AMD Versal, C++ inference)
I care deeply about:
- 📦 Reproducibility
- ⚙️ Systems design
- 🚀 Deployability, not just papers
- 📉 Latency, memory, and real constraints
- Raster & satellite imagery processing (SAR, optical)
- Patch extraction, tiling, reprojection, normalization
- Large-scale dataset engineering for DL pipelines
- UNet / GAN architectures for segmentation & translation
- End-to-end geospatial ML workflows (data → model → deployment)
Domains: Forest cover mapping, SAR image translation, land-use classification, large-area inference
- CNNs, UNet, GANs for dense prediction tasks
- Training on 10k+ satellite/SAR samples
- Evaluation under noisy, real-world conditions
- Model optimization for inference speed & memory
- Fine-tuning LLMs (LLaMA-2, T5)
- Retrieval-Augmented Generation (RAG)
- Document intelligence & semantic search
- AI agents for finance, sales, and analytics
- FastAPI-based inference services
- Python & C++ production code
- FastAPI microservices for ML inference
- Dockerized ML pipelines
- CI/CD, experiment tracking, reproducible training
- Bridging research → HPC → edge deployment
- Performance-optimized inference on AMD Versal VCK190
- C++-based pipelines balancing accuracy & latency
- Hardware-constrained deployment thinking
- GPU → FPGA/edge migration workflows
-
SAR Image Translation (GAN-based)
Operational deep learning models trained on large SAR datasets
(ISRO — production deployment) -
Forest Cover Classification (UNet)
Large-scale segmentation using Sentinel imagery
👉 https://github.com/ysant77/Forest-Cover-Classification
-
SemanticForce — Financial Document Intelligence
RAG-based system for querying and summarizing financial documents
👉 https://github.com/ysant77/PLP-PM-SemanticForce -
Investor Intelligence Agent
Automated investment analysis with neuro-fuzzy risk modeling
👉 https://github.com/ysant77/ISA-PM-InvestorIntelligenceAgent
- AI-Based Health Monitoring & Nutrition Planning
NLP-driven chatbot with predictive ML backend
👉 https://github.com/ysant77/AI-Based-HealthMonitoring
“Models are easy. Systems are hard.”
I optimize for:
- Clear ownership of data → model → deployment
- Designs that survive scale, noise, and change
- Clean abstractions with real performance numbers
- Practical AI that ships
Languages
Python · C++ · C#
AI / ML
PyTorch · TensorFlow · UNet · GANs · LLMs · RAG
Geospatial
Raster data · Satellite imagery · SAR · Remote sensing pipelines
Systems & MLOps
FastAPI · Docker · CI/CD · HPC workflows
Edge & Hardware
AMD Versal VCK190 · Performance-optimized inference
Cloud
AWS · GCP · GPU infrastructure
If you’re working on:
- Applied AI / ML systems
- Geospatial & remote sensing
- Robotics, perception, or edge AI
- Research-to-production ML
📩 Reach out on LinkedIn or explore the repos here.
I’m always open to:
- Research collaborations
- High-impact engineering roles
- Interesting problems that matter
⭐ If something here resonates with you, feel free to star a repo or start a conversation.