Back to home

A to Z

The longer version

How I got into applied AI, what I have shipped, and what I am still working on.

How I got here

I got into AI/ML during my CS undergrad at AIUB, initially hooked by how simple statistical models could pull clear signals out of noisy, unstructured data. But as much as I enjoyed the theory, I kept getting pulled toward a pragmatic question: what actually happens when this code meets real human workflows?

Publishing research on phishing detection at Tech Wing's Lab confirmed it for me—I loved the problem-solving, but I cared far more about building reliable systems than chasing incremental accuracy on static datasets. I wanted to see software fix actual operational headaches.

That led me to QualFix, where I moved from research into production engineering. These days, I split my time between building LLM agents, designing clean data pipelines, and architecting ERP systems—focusing on software that quietly replaces tedious manual work and holds up under real-world scale.

Education

BSc in Computer Science and Engineering

American International University-Bangladesh (AIUB)

CGPA 3.87 · Completed September 2026

Dean's List Honors, Faculty of Science and Technology: Fall 2023-24 (GPA 3.96) and Fall 2024-25 (GPA 3.95). Coursework: Artificial Intelligence, Machine Learning, Data Science, Data Analytics, SQL.

Work history

QualFix

Dhaka, Bangladesh

Web development and AI agency, Odoo partner

Technology and Business Lead

Current

August 2026 - Present

  • Own product strategy and technology direction for the company.
  • Scope and price client solutions, including quotes comparing an Odoo ERP rollout with a custom build.
  • Built an import-to-cash Odoo demo (landed costs, lot tracking) for a prospective food-import client.
  • Built the company's internal Odoo 19 ERP (CRM, sales, invoicing, projects, helpdesk) with nine custom modules, deployed on self-hosted Docker infrastructure via Dokploy across two legal entities.

AI Engineer (Full-time)

February 2026 - July 2026

  • Built an AI agent in LangGraph that reads incoming sales emails and their PDF, image and spreadsheet attachments and turns them into structured records in a single Gemini 2.5 Pro call, with the output schema enforced by Pydantic. It automates work previously done by hand.
  • Wired up the observability stack around it: MLflow 3.x for per-email traces and token cost, Prometheus + Grafana for queue depth, p95 latency, and monthly billing. Runs on FastAPI and async PostgreSQL inside Docker Compose.
  • Led an AI agent-assisted migration of 14 years of historical business data from Salesforce into Odoo, resolving schema differences between the two systems; completed in 4 weeks.
  • Built a separate POC for voice agents handling inbound and outbound calls; currently in review.

AI Engineer Intern (Remote, part-time)

July 2025 - January 2026

  • Raised an internal customer-matching system's accuracy from 37% to about 88% by reworking the matching logic and cleaning the underlying data.
  • Cleaned and structured contact data for medical organisations before loading into Postgres for downstream queries.

Tech Wing's Lab

Remote · Dhaka, Bangladesh

AI/ML Researcher, Team Lead (Part-time)

May 2025 - June 2025

  • Led the team's AI/ML research initiatives and collaborative projects.

AI/ML Researcher (Apprenticeship)

December 2024 - April 2025

  • Co-authored a peer-reviewed paper on phishing URL detection (ICERIE 2025). The best model, Random Forest with TF-IDF and Count Vectorizer features, reached 98.57% accuracy on the imbalanced dataset.
  • Drafted deep-learning architectures (RNN/LSTM) for real-time detection as a follow-up direction.

Research

Hybrid Machine Learning Framework for Phishing URL Detection with Advanced Feature Engineering

S. Datto, M. L. Kabir, et al. · Proc. 8th ICERIE 2025

Atlantis Press (Springer Nature), pp. 622-628, 2025 · Peer-reviewed

Best model reached 98.57% accuracy: Random Forest with TF-IDF and Count Vectorizer features, on an imbalanced dataset. Follow-up work drafted RNN/LSTM architectures for real-time detection.

Certifications

Industry Readiness Graduate Program (IRGP)

Lead Nation Academy, Batch 1

January 2026

HackerRank SQL (Basic & Intermediate)

December 2025

IBM Data Analyst Professional Certificate

May 2024

How I work

Reliability & cost visibility first: A model or agent is only as good as the system around it. I focus on getting telemetry right early—monitoring token consumption, p95 latency, and traces so we know exactly how a system performs and what it costs in production.

Better data over complex models: Throwing larger LLMs or heavier architectures at a problem is rarely the first answer. When fixing our customer-matching system, jumping accuracy from 37% to ~88% didn't require a bigger model—it came down to restructuring dirty inputs and rethinking the matching logic. Clean schemas beat raw model size almost every time.

Looking ahead: Over the next few years, I want to keep building at the intersection of AI engineering and business strategy—shipping practical systems that handle complex data, streamline operations, and deliver clear commercial value.

Outside work

When I'm not writing code or configuring infrastructure, I enjoy tinkering with self-hosted tools, keeping up with open-source AI projects, and taking long rides around Dhaka. I'm also passionate about community initiatives like local environmental cleanups.

Get in touch

Email is the best way to reach me: lutfulkabir1757@gmail.com

Technology and Business Lead @ QualFix, building AI agents, data pipelines and ERP systems that replace manual work.

lutfulkabir1757@gmail.com

Address

Dhaka, Bangladesh

lutfulkabir1757@gmail.com

+8801571337605

© 2026 Lutful Kabir