Independent researcher

Nabil Islam

Researching human-inspired AI.

I study how machines might reason, learn, and behave more like people, working across probabilistic reasoning, LLM architecture, and the research infrastructure beneath them.

LLMs Human-inspired AI Probabilistic reasoning Research infrastructure
Read my research Google Scholar ORCID
About

How boring am I, really?

A short introduction, for the curious.

I am an independent researcher focused on human-inspired AI: reasoning, efficiency, persona, feelings, self-estimation, and self-learning. One question started it all: do humans really have feelings, or do we learn them from our surroundings? It is hard to answer, but I hope to see AI develop something resembling feelings one day.

Most of my work is theoretical, which may sound dull. I enjoy designing a system's architecture and working out how its parts fit together.

Earlier, I worked in cybersecurity and cryptography: email-bombing attacks, infinite key-driven ciphers, and end-to-end messaging. I also build frameworks and research infrastructure. I believe knowledge should have no boundaries: “Learn everything. Satisfy the cravings of your brain.” When I eventually have a laboratory of my own, I plan to work across other disciplines too.

Research

What I investigate.

My work spans several disciplines, all pointed at one goal: AI that reasons and behaves more like us.

AI & LLM systems

Model orchestration, retrieval, local inference, fine-tuning, and efficient use of context.

Probabilistic reasoning

Property-based evidence aggregation and directional, multi-valued reasoning.

Knowledge infrastructure

Scholarly metadata, citation representation, retrieval, and machine-readable research data.

Cryptography

Experimental ciphers and applied approaches to secure messaging.

Cybersecurity

Attack analysis and the design of defensible, secure systems.

Publications

Papers.

My research so far. The list is short, and it is growing.

Loading papers…
Open source

Open-source work.

Where research and projects become reproducible and open to everyone.

Agent skill

BeHuman

Helps AI think and write more naturally, reducing obvious AI tells while preserving reasoning quality.

View on GitHub
Dataset

Scholarly Metadata Corpus

Structured scholarly metadata, collected for retrieval and local-model experiments.

View on Hugging Face
Standard

Object Rating Standard

An ESRB-style rating standard for software, websites, books, and games.

Visit the website
Python package

JuCrypt

An experimental symmetric cipher whose key is text you write yourself: a story, a sentence, a memory.

View on PyPI
Security tooling

Red-Scripts

Python scripts for vulnerability checking and security experimentation.

View on GitHub
Education

Academic background.

Formal education is only the starting point. I keep learning from everyone I meet.

2024

Motijheel Government Boys' High School

High School · GPA 5.00

2022

Rajdhani Ideal College

Secondary School · GPA 5.00

2019

Rajdhani Ideal College

Junior School · GPA 5.00

2016

Rajdhani Ideal Kindergarten

Primary School · GPA 5.00

Skills

Capabilities.

Tools I use daily. AI can do much of this too, which is why the research is the part I care about most.

AI & research

LLMsRAGLoRAPromptingEvaluationDatasets

Software

PythonFastAPIJavaScriptGitREST APIsAutomation

Systems & security

CryptographyCybersecurityLinuxNetworkingData formatsWeb systems
Credentials

Selected certificates.

Certificates are only worth what you build with them. These are a few I have earned along the way.

Loading certificates…