Business Growth Leader

Arif Adito.

Scaling startups into market leaders across SaaS, OTT, and Fintech with 15+ years of strategic impact.

Portrait of Arif Adito

I run Bangladesh for tapmad β€” building an OTT business from 0β†’1 in one of the world's most price-sensitive streaming markets: go-to-market, telco and payment partnerships, and the recurring-revenue engine behind them.

Fifteen years across OTT, SaaS, Fintech and Telecom β€” UAE, MENA, APAC and the UK. And I build with AI rather than talk about it: I pretrained a Bangla-focused language model end-to-end in Rust for $164, then published what broke.

Dhaka, Bangladesh

Explore
Track Record

15+ years of strategic impact.

A track record of scaling businesses from zero to market leaders across diverse industries.

Innovation & Ventures

Building the future.

Forging paths in technology, creative excellence, and open-access AI systems research.

Venture

Adioris Tech Ltd

A global technology powerhouse specializing in AI-driven solutions, digital transformation, and strategic consultancy. Bridging international expertise with local innovation.

AI SolutionsDigital TransformationStrategic Consulting
Research

LLM & Systems Research

Two arXiv papers: pure Rust end-to-end LM pretraining (arXiv: 2609.25008) and identity-anchored LLM architectures (arXiv: 2604.00009). Exploring frontiers in ML framework failure modes & cognitive modeling.

Pure Rust LM PretrainingFramework Failure TaxonomyIdentity-Anchored LLMs
Portfolio

What I'm building.

Everything I'm building β€” status pulled live from GitHub and Hugging Face.

● In DevelopmentFlagship

Noor

From-scratch Bangla-first hybrid language model β€” linear attention + mixture-of-experts, trained end-to-end.

$104 pretrain + SFT
private repo Β· v2 in training
● In DevelopmentFlagship

Eyla AIOS

Offline, Bangla-first agentic AI operating system that runs fully on consumer hardware.

16GB runs on a laptop
● Shipped Β· MIT

Gradient-Flow Arbiter

Verification harness that catches silent ML training bugs before compute is spent.

● Published Paper

Eyla Identity Paper

Identity-anchored LLM architecture with the Identity Consistency Score (ICS) benchmark.

● Research

AdiTurbo

Low-bit model quantization engine β€” sub-3-bit models that stay coherent.

● Shipped

Eyla Tool-Calling Models

On-device English / Bangla / Banglish tool-calling fine-tunes.

● Research

ADI

Predictive AI with a closed observe β†’ predict β†’ correct loop.

private betaβ€”
● MVP

MeshGPU

Peer-to-peer GPU rental marketplace.

private repoβ€”
● Shipped Β· production

Tapmad Anti-Piracy

OTT content-protection platform running in production.

Published Research

Written here, not gatekept.

Long-form research published directly, with every source listed and every unverified claim marked as such. No preprint queue, no paywall.

October 2026Β·10 sources

Why Long-Running AI Agents Fail: Cluttered Context or Compounding Mistakes?

AI agents fail on long jobs far more than their per-step accuracy suggests. Two explanations compete: context degradation (clutter in the model's context) and error compounding (one wrong step makes the next likelier). Context degradation is well measured across many models; compounding has strong theory but almost no direct measurement in LLM agents. A 16-run pilot on our own research pipeline shows a U-shaped failure pattern and a fresh-session run that never stalled. That points the same way, but it is underpowered and labelled as such. The piece ends with five fixes that help whichever mechanism dominates.

AI AgentsLong ContextReliabilityField Study
September 2026Β·62 sources

What Can Run Frontier-Competitive AI on a Low-Power CPU?

Eleven non-mainstream paradigms are surveyed against a single admissibility rule: a measurement counts only if it names the hardware, names the task, and reports energy rather than latency. Three carry admissible CPU evidence β€” Tsetlin Machines, retrieval-dominant tiny models, and HDC/VSA via the NystrΓΆm kernel bridge. The other eight fail as hardware-conditional, rebranded, or simply unmeasured. The field-wide finding is an energy blind spot that hits the winners as hard as the traps.

Edge AIModel EfficiencyBenchmarkingSurvey
August 2026Β·85 sources

Privacy Gateways and PII-Redaction Layers for AI Coding Agents

Five independent implementations already solve redaction-and-restore for Claude Code’s wire format. None of them combine local-first operation, reversible tokenization, offset-exact restoration, agent-aware coverage and regulated-vertical compliance in one product. This maps the landscape, the regulatory forcing functions across Bangladesh, Pakistan, India and the EU, and the four conditions under which the gap is real.

AI PrivacyCoding AgentsRegulatoryMarket Analysis
Continuous Learning

Professional Certifications.

40+ credentials across product, marketing, AI, data, and leadership β€” verified on LinkedIn

Claude Code 101

Aug 2026

Anthropic

Agentic coding with Claude Code β€” Anthropic Academy

New

Google Project Management: Specialization

Dec 2024

Google

6 course certificates incl. Agile PM & Capstone

Google Digital Marketing & E-commerce

Feb 2025

Google

Google Business Intelligence

Feb 2025

Google

Meta Certified Digital Marketing Associate

Feb 2025

Meta

Leading People and Teams

Feb 2025

University of Michigan

Web3 & Blockchain Leadership for Transformation

Jan 2025

INSEAD

Introduction to Generative AI

Dec 2024

Google

Salesforce Sales Operations

2024

Pathstream

Media & Recognition

Making headlines.

Recognition and coverage of market expansion milestones and strategic partnerships globally.

The Express Tribune

Pakistan's Tapmad enters Bangladesh market with Grameenphone

Coverage of Tapmad's strategic entry into the Bangladesh market through a major telecom partnership.

Read Article

PhoneWorld

Tapmad strengthens regional presence through partnership with Robi Axiata

Highlighting the expansion of Tapmad's footprint in South Asia.

Read Article

SAMENA Council

Pakistan's Tapmad expands to Bangladesh in partnership with Grameenphone

Industry recognition of the cross-border strategic collaboration.

Read Article
Get In Touch

Ready to build the future.

Currently exploring high-impact leadership roles in UAE, MENA, and global markets. Let's catalyze your next phase of growth together.

Book a call

Growth strategy, OTT & telecom GTM, or AI-native operations. Send a few details and Arif replies to confirm a time.

Book a call with Arif
ArifAdito.

Β© 2026 All Rights Reserved