Hi, I’m Nazrul 👋

I am a software engineer and ML engineer in New York City. I like building systems that have to be right: ML pipelines that replace manual checking, backends that thousands of people depend on, and LLM agents that call tools without guessing. I recently finished my MS in Computer Science at Oklahoma State University.

I am open to Software Engineer, ML Engineer, and AI Engineer roles.

70% → 93%LLM agent task completion
45× fasterML document validation
80,000+citizens served by APIs I built

Download Résumé Email Me GitHub LinkedIn

Featured Projects

Research · Oklahoma State University · 2024–2026

An LLM Agent That Calls Tools Without Guessing

LLM agents fail quietly: the model leaves out a parameter or invents one, and the tool still runs, against the wrong backend or for the wrong user.

What I built: a chat assistant where a LangGraph agent operates a zero-knowledge proof system through 8 MCP tools I wrote. I added 20 server-side checks that fill in or override what the model sends, and tested everything with my own 82-query suite across 3 model families (6,000+ runs).

Result: task completion rose from 70% to 93%, and a silent bug that sent 198 status checks to the wrong backend disappeared.

PythonLangGraphMCPFlaskRedisPostgreSQLDocker

Code Paper Architecture

The Prover Assistant chat: a user checks whether their data was used, then follows a proof from submission to verification

Industry · Tirzok Private Ltd. · 2023

ML Document Validation for a Mortgage Platform

Every mortgage application comes with a stack of borrower documents, and someone had to open and check each one by hand.

What I built: I led an ML pipeline for CliQloan, a mortgage loan origination system: classifiers sort each document and trim irrelevant pages, AWS Textract pulls out the key fields, and validation APIs check them automatically.

Borrower documents→Classify type→Trim pages→AWS Textract→Validate fields

Result: 45× faster than manual validation, with lower API costs.

Pythonscikit-learnAWS TextractLambdaECRSpring BootPostgreSQLAmazon MQ

Details

Research · Oklahoma State University · 2024–2026

Automatic DDoS Defense for Kubernetes

One flooded microservice can take down a whole app, and a service mesh like Istio only offers fixed rules that someone writes by hand.

What I built: a single monitoring pod that reads each service's traffic from its Envoy sidecar, detects attacks with a separate ML model per service, and patches Istio routing to move traffic to a backup version, then restores it when the attack stops. No application code changes.

Result: 94.0–97.8% detection accuracy, and parallel inference made each detection cycle 55.9% faster (37.4 ms vs. 84.8 ms).

Pythonscikit-learnKubernetesIstioEnvoyPrometheus

Prototype code Paper

How the DDoS detection and rerouting framework works

Industry · Tirzok Private Ltd. · 2023

Backend for a Government–Citizen Portal

A national portal that connects citizens of Bangladesh with government services has to stay fast and give each type of user the right access.

What I built: REST APIs and JWT-based role management for Janatar Sarkar, and I optimized its MongoDB aggregation queries.

Result: serves 80,000+ citizens, with 40% lower query latency.

Node.jsMongoDBJWTREST APIs

Project · Technical lead and sole architect · 2026–present · MVP

FwdStar: A Freight Marketplace, from Spec to MVP

In Bangladesh, freight is still booked offline between shippers, brokers, and carriers.

What I built: I turned the client's verbal needs into a technical specification and built the platform in nine phases: phone-first sign-up, fleet management, a load board, awards, GPS shipment tracking, and SMS alerts, with an English and Bengali frontend.

Result: 50+ documented REST endpoints and 20+ PostgreSQL tables, with row-level locking so a load is never awarded twice.

Next.jsTypeScriptFastAPIPostgreSQLPostGISDocker

Details

Smaller builds, including a LangChain + MCP chatbot and a license-plate reader, are on the Projects page.

Experience

  • Graduate Research Assistant, Oklahoma State UniversityJan 2024 – May 2026Built the LLM agent and the Kubernetes DDoS defense above, and contributed a merged fix to the open-source CoSMeTIC framework.
  • AI Instructor, NSF Research Experiences for Teachers, Oklahoma State UniversityMay 2025 – Aug 2025Taught Python, machine learning, and Transformers to 10 K–12 teachers.
  • Junior Software Engineer, Tirzok Private Ltd.Feb 2023 – Dec 2023Led ML document validation for CliQloan, 45× faster than manual checks; built APIs for a government portal serving 80,000+ citizens.
  • Undergraduate Research Assistant, BRAC UniversityOct 2020 – Jan 2023Built a cloud image retrieval framework that cut user waiting time by up to 54% (IEEE Transactions on Cloud Computing, corresponding author).

Full Résumé

Skills

LanguagesPythonJavaTypeScriptJavaScriptSQLBash
AI & MLLangGraphLangChainMCPPyTorchscikit-learnLIME
BackendFastAPIFlaskSpring BootNode.jsNext.jsREST APIsJWT
Cloud & InfrastructureDockerKubernetesIstioEnvoyPrometheusAWS
DataPostgreSQLPostGISRedisMongoDBRabbitMQ

Publications

A first-author paper on the LLM agent above is under review at IEEE BigData 2026, and I am the corresponding author of a paper in IEEE Transactions on Cloud Computing. All publications

Education

  • M.S. in Computer Science, Oklahoma State University · 2026 · GPA 3.82/4.0
  • B.S. in Computer Science, BRAC University · 2022

Outside Work and Contact

I am the founding captain of Lagaan, a cricket team in Stillwater, Oklahoma, which has won two trophies (Hobbies).

Email Me LinkedIn