Skip to content
ASK.
Arlington, VA

ABHINAYSAI

Building AI/ML systems that ship to real users.

Research · Engineering · Product

Worked at · Studied at · Certified by

Jio Platforms
George Washington University
Follett Higher Education
PHN Technologies
TEDx
Google Developer Club
AWS AI Practitioner
Global Leaders Award
Red Hat Certified
Google Advanced Data Analytics
Jio Platforms
George Washington University
Follett Higher Education
PHN Technologies
TEDx
Google Developer Club
AWS AI Practitioner
Global Leaders Award
Red Hat Certified
Google Advanced Data Analytics
What I build
VISION

Computer Vision

3D object detection on cryo-electron tomography. mAP@50 of 0.948. Built for structural biologists who need results they can trust.

SIGNALS

Time-Series + Signals

Benchmarking clinical EEG across 15+ architectures on 916 hours of real patient data. Honest evaluation over paper metrics.

MLOPS

MLOps + Infrastructure

The infrastructure that ships models and keeps them running. 100+ Airflow DAGs, Docker, K8s, Terraform, AWS/Azure/GCP. Observable at 3am.

AGENTS

Agentic AI

AI agents that route work between local and frontier. Privacy where it matters, reasoning where it counts.

Systems that shipped.

End-to-end projects in computer vision, clinical signal analysis, MLOps, and agentic AI. Each one grounded in honest evaluation. A real user at the end.

Tools of the trade.

What I actually use

ML & AI
PyTorchTensorFlowHugging Facescikit-learnLSTMMambaTransformersMoECenterNetYOLOv10Faster R-CNNMLflow
MLOps & Infra
DockerKubernetesApache AirflowTerraformJenkinsGitHub ActionsPrometheusGrafanaVaultCI/CDIaC
Cloud
AWSAzureGCPSageMakerEC2 + S3Databricks
Data Engineering
Apache SparkSpark StreamingApache KafkaBigQueryPostgreSQLMongoDBdbt
Languages
PythonSQLGoBashRTypeScript
Signal Processing
MNEEDF parsingTime-series windowingCLAHENumPyPandas

Rigorous engineer with taste.

MS in Data Science at George Washington University. Global Leaders Award, 3.77 GPA, graduating May 2026. Before grad school: ML infrastructure at Jio Platforms. Before that: biomedical imaging, clinical EEG, agentic systems. Same thread throughout. Honest evaluation, production-ready engineering, a real user at the end.

I care about the parts most engineers skip. Containerized training. Version-pinned environments. Observability that works at 3am. Infrastructure-as-code so the experiment you ran six months ago still runs today. These are what separate a notebook model from a system someone actually depends on.

Right now: finishing a seizure detection benchmark across 15+ architectures on 916 hours of pediatric EEG. Building an agentic research assistant with tool-use and retrieval. Preparing for what comes after May 2026.

0.948

mAP@50 — BYU cryo-ET

916

Hours of clinical EEG

94%

Pipeline reliability

85%

Faster model deploys

Research

Define the intelligence.

Grounding complex algorithms in robust logic, reproducible experimentation, and uncompromising mathematical evaluation.

Engineering

Scale the execution.

Building the automated, self-healing cloud architectures that allow models to survive contact with real-time traffic.

Product

Deliver the experience.

Bridging the gap between an engineering breakthrough and a tangible solution that solves a real human problem.

Where I’ve shipped.

May 2025 — Present

Washington, DC

Sales Operations Analyst

Follett Higher Education Group

  • Operate fulfillment and inventory data workflows during peak academic cycles. Online and on-campus. Maintaining process-level data integrity for hundreds of SKUs per cycle.
  • Apply CourseTracks adoption analytics for demand planning, check-in accuracy, and inventory reconciliation across multiple enterprise data sources.
  • Adhere to PCI-compliant processes and IAM-aligned access control. Zero security incidents across all operational periods.

Jul 2023 — Jul 2024

Navi Mumbai, India

Data Operations Engineer

Jio Platforms Limited

  • Engineered CI/CD automation for ML deployment across AWS/Azure/GCP using Jenkins, Docker, Kubernetes. Cut time-to-production 85% and release failures 60%.
  • Deployed Kubernetes-based ML inference microservices, reducing model-serving downtime 40% and enabling auto-scaling under real-time traffic.
  • Built 100+ Airflow DAGs with Vault-secured auth for ML and data workloads, achieving 99.9% reliability across multi-terabyte daily volume.
  • Optimized Spark Streaming for real-time feature pipelines, cutting data latency 40% under high-velocity workloads.
  • Provisioned cloud infrastructure via Terraform on AWS and Azure, achieving 30% cost reduction through right-sizing.
  • Integrated Prometheus + Grafana observability with automated alerting, cutting MTTD 45%.

Mar 2023 — Jun 2023

Pune, India

Data Analyst

PHN Technologies

  • Engineered data pipelines in BigQuery achieving 30% efficiency gains and 40% latency reduction. Enabling faster ML feature computation for downstream model serving.
  • Built 10+ real-time monitoring dashboards surfacing data quality issues and model drift signals. Enabling the team to catch performance regressions before they reached users.
  • Implemented key-based authentication in Airflow, securing 75% of critical data and ML pipelines while maintaining CI/CD velocity.

Certifications

  • AWS Certified AI PractitionerAmazon Web Services
  • Google Advanced Data AnalyticsGoogle
  • Google Business IntelligenceGoogle
  • Red Hat RHEL Automation with Ansible (RH294)Red Hat
  • Red Hat System Administration II (RH134)Red Hat

What I'm up to.

Updated May 2026

BuildingAgentic research assistant with tool-use + RAG
FinishingEEG seizure detection paper draft for publication
Open toCollaborations at the research-to-product boundary
ReadingThe Pragmatic Programmer (re-read for taste)
Based inArlington, VA

Let’s build

something real.

Open to research collaborations, product-ML builds, and teams working at the research-to-product boundary. If the problem is hard and the work is honest. Reach out.

Report an issue

Found something broken?

Layout bug, dead link, typo, or a suggestion — send it over. I read every one.

Sends via your email client. No tracking.