AI engineer and researcher · Rajshahi, Bangladesh

I build AI agents and study how they fail.

I lead AI engineering at an early-stage legal-AI company, where I build agent workflows for legal research, drafting, and review. Before that I led the Core RAG & AI team at AskTuring.ai, scaling its platform from 100 to 10,000+ concurrent users.

I also build open-source agent infrastructure and research agent security and alignment for my MSc at RUET.

Selected workResearch
Now
Co-founder & Lead AI Engineer, stealth legal-AI company
Before
Core RAG & AI team lead, AskTuring.ai
Research
Agent security and alignment, MSc at RUET
Open source
Golem, Herbie, LLM From Scratch
Awards
Champion, UCLA AltruAgent 2026 · Winner, Google Nano Banana 2025
Selected work

Agent infrastructure, an open book, and a few products.

2025 – Present

Go · MCP

Golem

Go framework for typed, inspectable agents

2026 – Present

Go · React

Herbie

Open-source, self-hosted AI platform

2024 – Present

TypeScript · MCP

Axiom Wiki

Knowledge base an LLM keeps up to date

2025 – Present

Rust · Tauri

Skill Manager

Desktop app to manage agent skills across tools

2026 – Present

PyTorch · Python

LLM From Scratch

Free book on training LLMs end to end

2025 – Present

Python · Evaluation

Agent Bench

Benchmark comparing agent frameworks

2026

Python · Game AI

Quicksilver

Autonomous Pokémon VGC Doubles agentChampion

2024 – Present

Next.js · NestJS

CareerKor

AI career platform for candidates and hiring teams

2025

Gemini · Stability AI

MagicSpin 360°

A 360° view from a single photoWinner
ExperienceFull experience →

Four years shipping AI in production.

Apr 2026 – Present

Remote

Stealth legal-AI companyCurrent

Co-founder & Lead AI Engineer
Own the AI architecture: agent workflows for legal research, drafting, and review, plus retrieval, evaluation, and deployment.

Jul 2025 – Apr 2026

Remote

AskTuring.ai

Applied AI Engineer, Core RAG & AI Team Lead
Led the RAG platform from 100 to 10,000+ concurrent users, cut hallucinations by 98%, and built an evaluation suite that cut eval time by 99%.

Oct 2023 – Jul 2025

Remote

Sazim Tech

Machine Learning Engineer
Shipped private, air-gapped LLM deployments for enterprise clients and a multi-provider LLM architecture.

Sep 2022 – Sep 2023

Remote

Intelsense AI

ML Researcher & Engineer
Built conversational AI for finance and telecom clients and worked on Bengali speech recognition.
ResearchResearch page →

Why agents fail, and how to measure it.

Ongoing

MSc thesis, RUET

Alignment drift after fine-tuning

How narrow fine-tuning can shift a language model's behaviour well outside the task it was trained on (emergent misalignment), and which mitigations work. The work starts by reproducing a published result on a controlled pipeline before testing anything new.

arXiv preprint

2026

Cyber-Capable AI Agents: Vulnerabilities, Evaluation Containment, and Defensive Response

A structured review of what goes wrong when cyber-capable models are given autonomy: multi-step attack chains, privilege separation, provenance, and who can reach a responder when an evaluation escapes its sandbox.

Ongoing

Dataset

Jontro: agentic data for Bangla

A tool-use dataset for Bangla-language agents. Tool results come from deterministic simulators, so every trajectory can be replayed and checked, and a first-call diagnostic tests whether a model picks the right tool in Bangla at all.
WritingAll writing →

Recent essays.

Oct 6, 2026

12 min read

Golem: A Type-Safe AI Agent Framework for Go

Most agent frameworks are written for Python. I wanted one that felt like Go: typed, explicit, stdlib only, and easy to test. So I built Golem.

Sep 10, 2026

10 min read

Can We Really Make AI Safe?

Can we really make AI safe when we do not fully understand the model, and we are steadily giving it more access to the world?

Sep 6, 2026

14 min read

Fireworks AI ZDR: How to Set It Up Without Missing the Exceptions

Fireworks has a good zero-retention default for open-model inference. This guide covers the API settings, cache isolation, residency, and feature boundaries that decide whether that is true for your application.

Jul 14, 2026

6 min read

Where Should the AI Actually Go?

A model can be capable of doing a task and still be placed in the wrong part of the system. I learned this while building AI for a legal workflow.
Contact

Get in touch.

For engineering roles, research collaboration, or a project. I reply within a few days.

abubakar1808031@gmail.com

Rajshahi, Bangladesh · UTC+6