Building Arioron
Leading an AI-first technology company while remaining directly involved in its products, models, research and technical direction.
00 / PERSONAL ARCHIVE
DHAKA · BANGLADESH · 2026
SAFWAT
SHABIB
ROLE
Founder & CEO, Arioron
FIELD
AI · Software · Research
EDUCATION
Cambridge A Levels
BASE
Bangladesh
SCROLL
01 / CURRENT
Now.
AUGUST 2026
BUILDING
Arioron Classroom — an AI-first education project currently in development.
RESEARCH
Investigating when a language model becomes committed to an incorrect answer and how that commitment might be identified, probed or causally altered.
ACADEMIC
Completing Cambridge A Levels while continuing independent research, technical writing and engineering work.
COMPANY
Continuing development across Arioron's models, products, developer tooling and research ecosystem.
02 / PROFILE
Context.
FOUNDER · RESEARCHER · STUDENT
I work across company building, research and implementation — moving from questions to experiments, and from experiments to systems people can actually use.
I am Safwat Shabib, a founder, researcher and engineer based in Bangladesh. My central focus is artificial intelligence: understanding intelligent systems, building them, evaluating them and turning promising ideas into practical products.
I founded Arioron as the organization through which that work could happen. Its ecosystem spans AI products, model development, research, datasets, developer tooling and infrastructure.
My research direction increasingly centers on reasoning and reliability: not only whether a model is right or wrong, but what happens before the final answer and when a wrong trajectory becomes difficult to reverse.
Alongside technical work, I am completing Cambridge A Levels and have maintained a long-running record of academic and STEM participation, including four IHSB STEM Fest editions across roughly eight years, South Breeze School STEM Fests in 2024 and 2025, Owlypia, World Scholars Cup, IHSB IDS 2026 and organizing at IHSB STEM Fest 2026.
Asking harder questions
Interested in model reasoning, reliability, efficient AI, evaluation and evidence synthesis.
Building end-to-end
Moving between applications, model development, deployment, infrastructure and interface design.
ROLE
Founder & CEO, Arioron
FOCUS
AI · ML · Systems · Research
EDUCATION
Cambridge A Levels
BASE
Bangladesh
RESEARCH IDENTITY
ORCID
0009-0009-1715-910X ↗
PUBLIC ARTIFACTS
Models & datasets
Hugging Face ↗
COMPANY
Arioron
arioron.com ↗
CERTIFICATIONS
HP LIFE Certified
HP LIFE CERTIFIED
Entrepreneurship
Business fundamentals and entrepreneurial thinking
View Certificate ↗HP LIFE CERTIFIED
Business Planning
Strategic planning and business model development
View Certificate ↗
03 / SELECTED WORK
Built.
PRODUCTS · SYSTEMS · RESEARCH
01
Arioron
The company I founded to build AI systems, research, models, developer tools and products under one ecosystem.
↗
02
VexAI
AI assistant and model ecosystem spanning language models, reasoning, multimodality, evaluation and experimentation.
↗
03
OncoDetect
AI research focused on medical imaging and cancer detection, including models, data and evaluation artifacts.
↗
04
Inquire
AI search and answer system focused on semantic retrieval, source-backed responses and information access.
↗
05
Classroom
Current Arioron project exploring an AI-first education experience.
↗
06
VexAPI
Developer-facing infrastructure connecting Arioron's AI capabilities with applications and technical workflows.
↗
WORKING STYLE
Question → prototype → evaluate → refine → deploy
WHAT I VALUE
Useful systems, measurable experiments and technically honest results
04 / RESEARCH
Questions.
ACTIVE · PAST · EMERGING
The interesting question is often not whether an AI is right. It is when the system crossed the point at which it became committed to the wrong answer.
My research spans language-model behavior, reasoning, reliability, efficient AI, model evaluation and structured evidence synthesis. I prefer questions that can be tested, reproduced and connected to concrete artifacts.
Understanding when and how a model commits to a reasoning path.
Studying model behavior beyond aggregate benchmark scores.
Exploring capability and practical efficiency at compact scales.
Developing rigorous structures for synthesis and evaluation.
01
When does an LLM become committed to being wrong?
Active research investigating the earliest reasoning stage at which a language model becomes committed to an incorrect answer, including token-level and latent-level probing, resampling, truncation, causal patching and steering.
↗
02
Structured Meta-Synthesis
Research into structured evidence synthesis with emphasis on transparent methodology, comparison across studies and reproducibility.
↗
03
Vex-Amber-Fable
Small-language-model development within the Arioron ecosystem, exploring capability and efficiency at compact parameter scales.
↗
04
Models, datasets & research artifacts
Public artifacts spanning language models, mathematical reasoning, medical AI and evaluation.
↗
05 / COMPANY
Arioron.
FOUNDER & CEO
A technology company built around a simple idea: research, engineering and product development should be able to operate in the same loop.
Arioron is where research interests become products, systems and experiments.
06 / ENGINEERING
Systems.
STACK · METHODS · PRACTICE
Models
LLMs · small language models · transformers · PyTorch · Hugging Face · model evaluation · inference.
Build
Python · JavaScript · TypeScript · React · Next.js · Angular · Flask · Three.js · GSAP.
Deploy
Linux · Docker · Nginx · VPS · cloud infrastructure · DNS · SSL · deployment · self-hosting.
Method
Literature review · evidence synthesis · experimental design · evaluation · reproducibility · technical writing.
Ship
Product architecture · prototyping · interaction design · developer tooling · technical strategy.
Tools
Git · GitHub · Figma · Postman · Hugging Face · LaTeX · Linux tooling.
07 / PUBLICATIONS
Written.
RESEARCH · TECHNICAL WORK
01
RESEARCH PAPER
Structured Meta-Synthesis
Safwat Shabib · Prerona Arman
02
ACTIVE RESEARCH
LLM Reasoning & Error Commitment
A research program investigating the earliest reasoning stage at which a model becomes committed to an incorrect answer.
03
TECHNICAL PUBLICATION
Vex-Amber-Fable-2.0
Technical work around a compact language-model release within the Arioron ecosystem.
04
RESEARCH ARTIFACTS
AI reasoning & evaluation datasets
Public datasets and research artifacts supporting mathematical reasoning, medical reasoning and model evaluation.
08 / OPEN RESEARCH
Artifacts.
MODELS · DATASETS · HUGGING FACE
MODEL
VEX-AMBER
Fable
Compact language-model research and development within the Arioron ecosystem.
OPEN HUGGING FACE ↗
MODEL
VEX-AMBER
Mini
Compact-model experimentation focused on efficient AI development.
OPEN HUGGING FACE ↗
MEDICAL AI
MODEL
OncoDetect Titan
Medical-AI artifact associated with the VexAI-OncoDetect research team.
VIEW MODEL CARD ↗
DATASET
REASONING
Vex-Math-L1-100K
Public mathematical-reasoning dataset within the Arioron Hugging Face organization.
VIEW DATASETS ↗
PUBLIC FOOTPRINT
Public models and datasets provide a direct trail from research ideas to released technical artifacts.
09 / ACADEMIC & COMMUNITY ACTIVITY
Beyond.
PARTICIPATION · SERVICE · ORGANIZATION
My academic and community record extends beyond research and company building. I have participated in STEM and interdisciplinary programs over many years, and more recently moved from participation into service and organization.
2026
IHSB STEM Fest
Organizer · Hospitality
2026
IHSB IDS
Volunteer
2025
South Breeze School STEM Fest
Participant
2024
South Breeze School STEM Fest
Participant
~8 YEARS
IHSB STEM Fest
Participant · 4 editions
EARLIER
Owlypia
Participant
EARLIER
World Scholars Cup
Participant
IHSB STEM Fest
Participated in four IHSB STEM Fest editions across roughly eight years, beginning as a child and continuing through 2026. In 2026, that long-running participation progressed into an organizing role in Hospitality.
South Breeze School
Participated in the South Breeze School STEM Fests in 2024 and 2025.
Owlypia
Participated in an interdisciplinary academic environment centered on broad knowledge, reasoning and intellectual engagement.
World Scholars Cup
Participated in an international academic competition combining scholarship, communication, debate and collaborative challenges.
IHSB IDS 2026
Volunteered at IHSB IDS 2026, contributing to event operations and the wider school community.
IHSB STEM Fest 2026
Served as an organizer in Hospitality, contributing to the execution and event experience.
PROGRESSION
Participation → sustained involvement → service → organization.
10 / TIMELINE
Becoming.
2022 → NOW
The trajectory has moved from software and experimentation toward company building, model development and research.
2022
Early building
Beginning the journey through software, technology, experimentation and entrepreneurship.
2023
Innovex Int'l
Moving from individual technical work toward a broader company-building effort.
2024
Strategic reset
A period of restructuring and rethinking the organization's direction.
2025
Arioron
Expanding into AI systems, products, models, infrastructure and research under the Arioron identity.
2025–2026
Models & research
Vex-Amber, OncoDetect, public datasets, publications and a deeper focus on AI research.
2026
Classroom & reasoning
Building Arioron Classroom while developing research around LLM reasoning and reliability.
NOW
Next
Continue building, publishing, experimenting and turning difficult technical questions into concrete systems.