Senior Research Engineer @ SMU · Applied AI / ML Systems · LLMs · Remote Sensing · GPU / Edge AI
Singapore · NUS MTech — Intelligent Systems · Former Scientist/Engineer @ ISRO
yatharth@singapore:~$ profile
role Senior Research Engineer @ SMU, Centre for Digital Law
background Scientist/Engineer @ ISRO
focus LLM systems · legal AI · SAR / remote sensing · GPU / edge ML
stack Python · PyTorch · C++ · FastAPI · PostgreSQL · Docker · HPC
building reliable ML systems from research → deployment
learning GPU acceleration · quantitative ML
I build end-to-end machine learning systems across research, data engineering, model development, evaluation, and deployment.
My work sits at the intersection of AI research and systems engineering: LLMs and information retrieval, remote sensing and SAR, performance-aware ML, and production-grade infrastructure. I care about systems that remain reproducible, measurable, and useful outside a notebook.
Models are easy. Systems are hard.
| Area | What I work on |
|---|---|
| LLMs & Information Systems | RAG, document intelligence, retrieval, information extraction, evaluation, structured knowledge |
| Remote Sensing & SAR | SAR / optical imagery, temporal modeling, segmentation, translation, geospatial data pipelines |
| ML Systems | Training and inference pipelines, APIs, data engineering, experiment reproducibility, HPC workflows |
| Edge & Acceleration | Hardware-aware inference, C++, AMD Versal, latency / memory trade-offs, GPU-to-edge deployment |
Building reliable pipelines for extracting, grounding, retrieving, and validating information from complex legal documents, with emphasis on provenance and evaluation.
Large-scale Sentinel imagery segmentation with deep learning and geospatial preprocessing workflows.
Retrieval-augmented document intelligence for querying and summarizing financial information.
Automated investment-analysis system combining ML-driven reasoning with structured risk modeling.
Deep-learning work across SAR imagery, large raster pipelines, image translation, temporal modeling, and operational geospatial ML — including production-facing work at ISRO.
research → data → model → evaluation → system → deployment
Languages
Python · C++ · C#
ML / AI
PyTorch · TensorFlow · Transformers · LLMs · RAG · CNNs · UNet · GANs
Systems
FastAPI · PostgreSQL · Docker · Git · CI/CD · HPC
Geospatial
SAR · Satellite imagery · Raster pipelines · Remote sensing
Performance / Edge
C++ inference · AMD Versal VCK190 · GPU workflows · Hardware-aware ML
- Reproducibility — experiments and pipelines should be repeatable.
- Evaluation — model quality needs evidence, not vibes.
- Deployability — the system around the model matters as much as the model.
- Performance — latency, memory, throughput, and hardware constraints are first-class concerns.
- Clear abstractions — research code should have a path toward maintainable engineering.
$ current_focus
→ reliable LLM / retrieval systems
→ multimodal and temporal SAR modeling
→ GPU acceleration and efficient inference
→ quantitative ML and mathematical finance
If you're working on applied AI, ML systems, LLMs, remote sensing, performance engineering, or research-to-production ML, feel free to reach out.
LinkedIn: linkedin.com/in/yatharthsant