Available for meaningful problems

I build AI systems that hold up in the real world.

Kawsher HRidoy is an ML & Systems developer and Team Lead of Cortex Crew at Daffodil International University, Dhaka. He takes teams from a practical problem to a live, defensible system.

Explore selected work Email Kawsher
05
competition results
02
podium finishes
04
showcased systems
01
championship
( 01 )THE THROUGHLINE

The best AI work is not a demo. It is a dependable system that earns its place in somebody's day.

I lead teams from a messy problem to a live, defensible product - pairing applied machine learning with careful engineering, clear interfaces, and a bias for shipping.

( 02 )SELECTED SYSTEMSGitHub profile
01CYBERSECURITY

Stealer-malware victim awareness

Darktrace3

Makes the blast radius of an infostealer incident understandable without ever exposing the passwords it found.

The problem. Malware and infostealer logs can leave people unsure which accounts were put at risk and what to do next.

What we built. Cortex Crew turned exposed-site and account evidence into a clear awareness view, with passwords kept masked throughout.

Important boundary. Victim awareness requires useful evidence, not the unnecessary disclosure of sensitive credentials.

CSAD 2026 ChampionSource private
  • PYTHON
  • SECURITY
  • DATA VISUALISATION
02HEALTH SYSTEMS

Patient-owned medical records

Niro

Brings Bangla medical documents, time-series health records, and doctor-ready context back to the patient.

The problem. Lab reports, prescriptions, and medical history are difficult to keep together and difficult to interpret across visits.

What we built. A patient-owned record with Bangla document reading, time-series tracking, and an organised case summary for the doctor.

Important boundary. The AI helps organise context; it does not give the final medical advice. The clinician remains in the loop.

IEEE ICADHI 2026 1st Runners-upSource private
  • FASTAPI
  • BANGLA OCR
  • HEALTH DATA
03ML INFRASTRUCTURE

GPU failure prediction

Autopilot

Uses GPU telemetry to make time for recovery before a failing node takes a running job down.

The problem. GPU node failures interrupt valuable running work when the infrastructure has no time to react.

What we built. A telemetry-driven prediction loop that identifies potential failures about 100 seconds ahead, cordons the node, and migrates the running job.

Important decision. The prediction is tied to a concrete Kubernetes recovery action, so the signal has an operational use.

AI Innovation Hackathon 2026 FinalistView source
  • LIGHTGBM
  • KUBERNETES
  • GPU TELEMETRY
04EDUCATION SYSTEMS

Auditable mentor briefing

AI Mentor

Turns academic and attendance evidence into a transparent priority view and a useful briefing for mentors.

The problem. Mentors need to see who needs attention and why, without a black-box score becoming the decision.

What we built. A system that evaluates eight academic and attendance signals using deterministic rules, then adds an AI-generated mentor brief.

Important decision. Ranking stays deterministic and auditable; AI explains context for the person who makes the decision.

AI Project Competition 2026 FinalistView source
  • NEXT.JS
  • AZURE OPENAI
  • AUDITABLE RULES
( 03 )2026 COMPETITION RECORD

05

Five public results in 2026: one championship, one 1st Runners-up finish, and three finalist placements.

  1. CHAMPION

    CSAD 2026 Project Showcasing Competition, Cyber Security Centre, DIU

    Darktrace3
  2. 1ST RUNNERS-UP

    IEEE ICADHI 2026 Project Showcase, IEEE DIU Student Branch

    Niro
  3. FINALIST

    AI Innovation Hackathon 2026, Department of CSE, DIU

    Autopilot
  4. FINALIST

    AI Project Competition 2026, Department of CSE, DIU

    AI Mentor
  5. FINALIST

    5th Data Science Summit, DIU

    Niro
( 04 )ENGINEERING APPROACH
01

Deterministic where it counts

Use clear, inspectable logic for scoring, ranking, and actions that need to be defended.

02

AI explains; people decide

Put AI where it clarifies evidence and makes context usable, while keeping accountable humans in control.

03

Measure against the real baseline

Start with the failure, workflow, or user burden already present and make the technical decision answer to it.

( 05 )ABOUT & STACK

ML is only useful when the system around it is ready.

Kawsher HRidoy leads Cortex Crew from early planning through a live competition demo. His work joins model thinking, backend systems, product interfaces, and deployment realities - with a current focus on competitive cybersecurity.

Based in Dhaka, Bangladesh, and building with Cortex Crew at Daffodil International University.

MODELLING

Python LightGBM

BACKEND & DATA

FastAPI PostgreSQL

FRONTEND

Next.js TypeScript

INFRASTRUCTURE

Docker Kubernetes Linux

AI SERVICES

Azure OpenAI

( 06 )CONTACT

Have a hard problem? Let's make it useful.

kawsher@hridoy.xyz
GitHub LinkedIn Facebook Instagram