I recently finished my MS in Computer Science at Oklahoma State University in May 2026. For more than two years, I worked there as a Graduate Research Assistant with Dr. Paritosh Ramanan and Dr. Sharmin Jahan. With Dr. Ramanan in the Distributed Intelligent Systems Lab, I built an LLM agent that lets people use zero-knowledge proof systems in plain language, a project I am still working on. With Dr. Jahan, I built an ML system that detects DDoS attacks on cloud services and responds to them automatically.

Before my master’s, I spent more than two years as an undergraduate research assistant with Dr. Jannatun Noor at BRAC University, working on faster image retrieval for cloud systems and leading fieldwork in remote Indigenous communities of Bangladesh. I also worked for a year as a software engineer, building an ML document-validation system for a US mortgage-software company and the backend of a national government portal.

I am applying to Computer Science PhD programs for Fall 2027. CV · Publications · Projects · Google Scholar · GitHub

News

• Oct 2026: My pull request to the CoSMeTIC zkSNARK framework was merged, including a fix for a bug that reported every valid KS and LRT proof as failed.
• Sep 2026: Released the code, test suite, and all 6,371 result records from my skeptical-tools paper. [Code & data]
• Fall 2026: Three more papers under review, at ICOIN 2027, ACM CHI 2027, and Social Sciences & Humanities Open.
• Jul 2026: Our paper on Indigenous primary education was accepted at ACM COMPASS 2026. [Paper]
• May 2026: Completed my MS in Computer Science at Oklahoma State University.
• Summer 2025: Taught AI to 10 K–12 teachers as an AI Instructor in OSU’s NSF Research Experiences for Teachers program.

Selected Research

Making LLM Agents Reliable When They Run Zero-Knowledge Proof Systems

First author · with Dr. Paritosh Ramanan · under review, IEEE BigData 2026 · Paper · Code & data

Zero-knowledge proofs let a clinical study prove that its statistics are correct without showing any patient records, but requesting a proof means writing exact API calls, which shuts out the participants and auditors these systems are meant to serve. I built an LLM agent that lets people request, check, download, and verify proofs in plain language, using eight custom MCP tools on top of the CoSMeTIC zkSNARK framework. Because LLMs often leave out or invent the exact values these tools need, I designed skeptical tools: 20 server-side checks that complete or override what the model sends.

• Task completion rose from 70% to 93% on Qwen 3 32B, with a gain of about 23 points on all three model families we tested (6,000+ query runs).
• A “plausible default” inside one tool silently sent 198 status checks to the wrong prover, and the models then told users their job did not exist. A server-side override removed all 198 errors.
• Made-up answers appeared only when the chat history contained an earlier success to copy.
• The code, test suite, and all 6,371 result records are public, and I contributed a merged fix to CoSMeTIC itself.

Automatic DDoS Defense for Cloud Service Meshes

Second author · with Dr. Sharmin Jahan · under review, ICOIN 2027 · Paper · Prototype code

Modern apps on Kubernetes are made of many small services, and a flood of requests to one of them can bring the whole app down, yet service meshes like Istio only offer fixed rules that people set by hand. I built a framework that runs as a single monitoring pod and needs no changes to application code. It reads traffic data from each service’s Envoy sidecar, detects attacks with a separate ML model for each service, and patches Istio routing to move traffic to backup versions, then restores normal routing when the attack stops.

• Per-service models (Random Forest, Gradient Boosting, SVM, Decision Tree) reached 94.0–97.8% accuracy on 4,705 labeled traffic windows.
• Running the models in parallel made each detection cycle 2.27× faster than one central model across 12 pods.
• My earlier prototype analyzed more than 1.2M Envoy log entries and explained each detection with LIME.

Other Research

The digital divide in remote Indigenous communities · Second author · under review, ACM CHI 2027 [Paper] and Social Sciences & Humanities Open [Paper] · related paper at ACM COMPASS 2026 [Paper]

We trekked for days through the hills of Bandarban, Bangladesh, to reach 15 villages of six Indigenous communities, many with no mobile network, electricity, or roads. More than half of the people we met had no mobile coverage at home, and many climbed to mountain peaks just to make a call. Our papers describe this “physical access” layer of the digital divide and how to do respectful fieldwork in such places.

Faster image retrieval over the cloud · Corresponding author · IEEE Transactions on Cloud Computing · [Paper] We customized progressive JPEG so images load faster and take less space in low-bandwidth settings, cutting user waiting time by up to 54% and image size by up to 27% in a real-world test across two continents.

DDoS mitigation with a resource-sharing network · Co-author and oral presenter · NSysS 2022 · [Paper] A network that tracks attacker IP addresses and drops their requests through a proxy, which mitigated 66.7% of attacks in a simulation with 50 Nginx virtual machines.

Industry & Engineering

• ML document validation for Kramasoft, a US mortgage-software company: classifiers that sort borrower documents and remove irrelevant pages before AWS Textract, running 45× faster than manual validation.
• Backend of Janatar Sarkar, a Bangladesh government portal serving 80,000+ citizens: RESTful APIs and a JWT-based role management system.
• FwdStar, a freight marketplace platform for Bangladesh (technical lead and sole architect, 2026–present): a Next.js and FastAPI system with 50+ REST endpoints, PostGIS, and layered security, built from a 2,695-line specification I wrote from the client’s needs.

Teaching and Beyond

I taught AI to 10 K–12 teachers as an AI Instructor in an NSF-funded program at Oklahoma State. Outside research, I am the founding captain of a cricket team in Stillwater, Oklahoma (Hobbies).