Hello World

I'mSehastrajit

>

Building scalable ML systems at the intersection of LLM inference, computer vision, and cloud-native pipelines. M.S. CS @ ASU · 4 published papers · shipped at scale.

View My Work
Contact
4+
Publications
89%
Model Accuracy
35%
Context Retention
30%+
Latency Reduced
scroll

01 / About Me

The Human Behind the Models

AI/ML Engineer pursuing a Master's in Computer Science at Arizona State University. My work spans LLM systems, large-scale computer vision, and cloud-native data pipelines — built to scale, built to ship.

I design end-to-end ML workflows: from data ingestion and feature engineering through model training, optimization, and production deployment. I care about measurable impact — not just accuracy metrics, but latency, reliability, and real-world adoption.

Formerly a Machine Learning Engineer at Omdena (where I pushed model accuracy from 72% → 89%), a Research Data Scientist at Auburn University, and a Research Aide at ASU's School of Sustainability building geospatial CV pipelines for solar infrastructure analysis.

ASU AI ScholarGoogle DSC MemberOpen Source ContributorIEEE Author
🧠
LLM Systems
Production inference stacks with vLLM, Triton, and quantization for 30%+ latency gains
👁
Computer Vision
Geospatial + temporal image pipelines, satellite imagery analysis, PV infrastructure detection
☁️
Cloud MLOps
AWS & GCP pipelines, Docker/K8s deployments, CI/CD workflows and MLflow experiment tracking
📄
Published Researcher
4 peer-reviewed papers across IEEE, Evolutionary Intelligence, and bioinformatics journals

Education

M.S. Computer Science
Arizona State University
Aug 2025 – Aug 2027
Statistical ML · Data Processing at Scale · Knowledge Representation
B.Tech. CS (AI & ML Specialization)
Vellore Institute of Technology
Jul 2021 – Jul 2025
Deep Learning · NLP · Image Processing · Probability & Statistics

02 / Experience

Where I've
Built Things

01 / 05
Research · Tempe, AZ

Research Aide

ASU School of Sustainability

Apr 2026 – Present
  • Built computer vision and geospatial pipelines for temporal photovoltaic (PV) infrastructure analysis across multi-year NAIP imagery
  • Automated preprocessing, alignment, and cross-year comparison of multi-resolution satellite imagery (0.3m–0.6m) for longitudinal infrastructure monitoring
  • Quantified image fidelity and spatial consistency using PSNR, SSIM, RMSE, and MAE to support large-scale sustainability research
Computer VisionGeospatialNAIP ImageryPythonPyTorch

03 / Projects

Things I've Shipped

End-to-end ML systems — from architecture design to production deployment and benchmarking.

🧠
Jan 2025 – May 2025

TITANS: Memory-Augmented LLM

Designed and deployed an optimized memory-augmented LLM inference stack with measurable retention and latency gains, inspired by the TITANS architecture paper.

  • Custom memory-augmented transformer achieving 35% improvement in long-context retention on BoolQ and GSM8K
  • Deployed optimized LLM inference stack using Triton, vLLM, and FastAPI for low-latency serving
  • Reduced inference latency by 30%+ through quantization, batching optimization, and benchmarking
vLLMTritonFastAPITransformersPyTorchQuantization
🛰
Aug 2025 – Present

Cooling Tower Detection from NAIP Imagery

Computer vision pipeline for detecting cooling tower locations and estimating physical dimensions from aerial NAIP imagery to support sustainability research.

  • Geospatial image processing pipeline for pixel-to-real-world spatial measurement conversion
  • Large-scale aerial imagery processing with automated preprocessing for infrastructure analysis
  • Integrated with ASU School of Sustainability's longitudinal monitoring research
Computer VisionGeospatialNAIPPythonCV Pipelines
📊
Jul 2024 – Oct 2024

Blue Zones Longevity ML Platform

Scalable ML analytics platform for analyzing longevity and lifestyle factors using vLLM, inspired by Netflix's 'Live to 100'. Led a team of 15 at Omdena.

  • 72% → 89% predictive accuracy improvement through feature engineering and pipeline optimization
  • 40% preprocessing latency reduction with structured logging over 1M+ records
  • Actionable public health insights for policy recommendations on longevity factors
vLLMML PipelinesFeature EngineeringData VizPython

04 / Skills

The Full Stack

From raw data to production-grade ML systems — across languages, frameworks, and cloud platforms.

PyTorchvLLMTritonFastAPIMLflowDockerAWSGCPTransformersComputer VisionGeospatialCI/CDKubernetesTensorFlowScikit-learnPythonSQLGit
{ }

Languages

Python95%
TypeScript80%
C/C++72%
Java68%
SQL82%

ML & AI

PyTorch92%
TensorFlow85%
Transformers90%
Scikit-learn88%
Computer Vision87%
Deep Learning90%
🚀

LLM & MLOps

vLLM85%
Triton Inference78%
FastAPI82%
MLflow80%
CI/CD75%
Model Versioning80%

Cloud & Systems

AWS (S3/EC2/Athena)82%
GCP (Vertex/BigQuery)78%
Docker85%
Kubernetes72%
Git90%

07 / AI Experience

Meet Luna

Groq · Llama 3.3 70B · Live

Not a generic chatbot. Luna is purpose-built for this portfolio — she generates pitches, matches job descriptions, writes cover letters, and answers anything about Sehastrajit. Powered by Groq's fastest inference.

Model
Llama 3.3 70B
Provider
Groq
Context
128K tokens
Tokens/sec
~800
0 chars

05 / Publications

Peer-Reviewed Research

4 published papers across IEEE, international journals in AI, medical imaging, and bioinformatics.

4
Papers Published
3
Journals
2024
Latest Publication
IEEE
Top Venue
2024Journal

Pipelined Structure in the Classification of Skin Lesions Based on AlexNet CNN and SVM Model With Bi-Sectional Texture Features

IEEE Access

Computer VisionCNNSVMMedical ImagingIEEE
2023Journal

Comparative Performance of Deep Learning Architectures in Classification of Diabetic Retinopathy

International Journal of Ad Hoc and Ubiquitous Computing

Deep LearningMedical AIRetinopathyCNNBenchmarking
2023Survey

An Insight on Recent Advancements and Future Perspectives in Detection Techniques of Parkinson's Disease

Evolutionary Intelligence

Parkinson's DiseaseML DetectionSurveyBiomarkers
2024Journal

The Estimation of Statistical Features from VMD Levels for Automated Sleep Apnoea Classification

International Journal of Bioinformatics Research and Applications

Sleep ApnoeaVMDSignal ProcessingBioinformatics

06 / Contact

Let's Build Something

Open to ML engineering roles, research collaborations, and interesting problems at the intersection of LLM systems, computer vision, and production AI.

Say Hello

sehastrajit@gmail.com · Greater Phoenix Area

<Sehastrajit /> · AI/ML EngineerDesigned & Built from scratch · 2026