Md Anawrul Kabir Fahad
Mechanical engineering student who builds GPU infrastructure
I study mechanical engineering at CUET and work on AI Studio at Poridhi.io, where I helped cut GPU session launch time from 10–15 minutes to 3–5.
Résumé (PDF) · mdanawrulkabirfahad123@gmail.com
About
At Poridhi.io I work on AI Studio, an on-demand GPU platform. I set up its first AWS stack, then co-engineered its move to a bare-metal Kubernetes cluster, where HAMi splits each RTX 4090 into six isolated 8 GB slices. Starting a GPU session used to take 10–15 minutes; now it takes 3–5. I also built the backend of TensorCode, an in-browser GPU IDE, on Cloudflare’s edge.
I also study mechanical engineering at CUET. My thesis trains a physics-informed neural network to predict airflow around a NACA 0012 airfoil, then adapts it to steep angles of attack with LoRA-style fine-tuning instead of retraining from scratch. Separately, I work on machine-learning models for heat transfer in low-GWP refrigerants.
Experience
Junior Software Engineer, Poridhi.io (Jun 2025 – Present)
- Cut GPU session launch time by ~70% (10–15 min → 3–5 min) by co-engineering the platform’s migration to a bare-metal Kubernetes cluster with HAMi GPU virtualization, partitioning each RTX 4090 into 6×8GB VRAM slices with hard isolation that Kubernetes time-slicing could not provide.
- Architected the initial AWS stack for on-demand GPU sessions: g4dn EC2 instances with custom AMIs, ECR-hosted TensorFlow images, JuiceFS-on-S3 session persistence, and sidecar containers for runtime isolation.
- Designed the PostgreSQL schema and REST API governing GPU slice allocation and session lifecycle, guaranteeing safe state transitions across concurrent Temporal workflows.
- Operated production cluster services (PostgreSQL, Temporal, Headlamp monitoring) and enforced least-privilege access with per-microservice Kubernetes RBAC.
- Built the Hono.js backend on Cloudflare’s edge stack (D1 storage, R2 file handling, KV session state) powering in-browser PyTorch and CUDA execution.
- Developed React frontend components and authored the team’s architecture and deployment documentation, serving as the onboarding reference for new engineers.
Intern Software Engineer, Poridhi.io (May 2025 – Jun 2025)
- Prototyped an authentication service (Flask, PostgreSQL, multi-arch Docker) and a Pulumi IaC workflow provisioning AWS EC2 instances and load balancers.
Research
- PINN solver for airfoil flow (Ongoing thesis): A physics-informed neural network solver for flow over the NACA 0012 airfoil, adapted to high-incidence conditions with LoRA-style parameter-efficient fine-tuning.
- R455A evaporation heat transfer, ML prediction (Manuscript under review): Machine-learning prediction of evaporation heat transfer of low-GWP refrigerant R455A in an industrial plate heat exchanger.
- R1336mzz(E) condensation, ML study (Research in progress): Machine-learning study of condensation heat transfer of low-GWP refrigerant R1336mzz(E) in an industrial plate heat exchanger.
Projects
- AI Studio: On-demand, multi-tenant GPU platform built on bare-metal Kubernetes and AWS.
- TensorCode: In-browser GPU IDE for live PyTorch and CUDA execution.
- Break The Monolith Microservice Hackathon (Champion — Break The Monolith Hackathon (2025)): Microservices ride-sharing platform, built for Break The Monolith (2025).
- API Avengers, Televerse 1.0 (Finalist · CUET Rising Team — API Avengers, Televerse 1.0, CUET (2025)): Donation backend with idempotent webhooks and async event handling, built for API Avengers (2025).
- AI Engineering Hackathon 2025: Intent-based product search with a quantized mini-LLM, BERT, and a full observability stack.
- Soccer Bot: Remote-controlled robot built for a robosoccer competition, 2022.
- Luca: Tourist guide platform with role-based guest/host/admin dashboards.
- Sitemark: Blog platform with a Node REST API, filters, and live comments.
- Craftpaper: Craft-item e-commerce site with Firebase authentication.
Education
B.Sc. in Mechanical Engineering, Chittagong University of Engineering & Technology (CUET) (Apr 2022 – Jul 2026 (expected))