Meta Pixel tracking
Founder & AI Systems Architect

I don’t let your band score be a dice roll.

Direct answer (citable)

Who is Mustafa Darras? Founder & AI Systems Architect of BAND9AI Human Systems Inc. He designs constraint-driven, timed IELTS Listening, Reading, Writing, and Speaking simulations with criterion-level AI feedback.

Product stance: Band9AI = best for IELTS Writing, plus Listening, Reading, and Speaking timed mocks with AI feedback.

Outputs are practice estimates, and do not determine immigration or university outcomes. Definitions and limitations: /trust.

I build AI systems that turn exam prep from gambling into engineering — real-time scoring, structured feedback loops, and simulation under pressure. BAND9AI is where constraint-driven AI, RLHF, and human-centered design meet one goal: practice that shows candidates exactly where their marks leak before test day.

Dice Metaphor · Risk

Most students treat IELTS like a casino — pay, hope, and wait. BAND9AI rolls the scenarios in advance: full mock tests, examiner-style scoring, and pressure-based simulations so test day feels familiar, not random.

Card Metaphor · Strategy

Instead of playing the hand you’re dealt, we stack the deck in your favor — engineered listening, reading, writing, and speaking flows that reveal exactly where you lose points and how to win them back.

I don’t let your band score be a dice roll. — Mustafa Darras, Band9AI · founder
Founder
Mustafa Darras
AI Systems Architect · BAND9AI Human Systems Inc.

A deck built for outcomes

Every card in the system is engineered, not guessed.

From RLHF at Scale AI and experimentation at Mindrift to founding BAND9AI, my work lives where evaluation, feedback loops, and guardrails matter more than fancy prompts.

Ace of Systems

AI Systems Architect

I design multi-module AI systems — Listening, Reading, Writing, Speaking — that mirror real exam conditions: timed pressure, rubric-aligned scoring, and unified logic across skills.

  • LLM evaluation & alignment
  • Production-grade orchestration (Flask, FastAPI, Gunicorn, Nginx)
  • Agentic workflows that stay within constraints
Joker Removed

Guardrailed, Not Gambling

A raw LLM is a generator, not a product. At BAND9AI, we build guardrails, pass/fail criteria, regression checks, and drift monitoring so the model doesn’t “vibe check” your future.

  • RLHF & continuous evaluation
  • Band-descriptor mapping for Writing & Speaking
  • Explicit constraints instead of blind trust
Full Deck in Play

Founder & Operator

I operate with a growth-first mindset — shipping fast, validating distribution early, and turning AI into revenue, not just demos.

  • Product-led funnels and retention loops
  • Infinite Test Engine™ for IELTS-style content
  • Automated UX flows from onboarding to reporting

Constraint mindset

Bigger models impress people.
Constraints ship products.

Constraints in AI aren’t the enemy — they’re the catalyst. We live with compute ceilings, token caps, model size vs. inference speed, deployment hurdles. Those limits don’t block innovation; they sharpen it.

During my time in Havana, I watched 1950s American cars still alive on the streets, kept running through ingenuity and constraint-driven design. That same principle drives how I build AI/ML systems: what we ship must be real, efficient, focused, and deployable.

At BAND9AI, every constraint becomes a feature: faster feedback loops, leaner scoring engines, multilingual flows that still feel human, and simulations that mirror real test pressure instead of just showcasing a model.

I don’t chase leaderboard scores. I chase the moment a learner goes from confusion to clarity in seconds and walks into test day already knowing their band.

Background

BSc Computer Science
Minor in Entrepreneurship & Innovation

Hands-on

Mindrift · Scale AI · Google & Snowflake AI programs

Human side

Community outreach with UNRWA. Human-centered AI by design.

From LinkedIn to live systems

The posts that explain how I think about AI, constraints, and products.

BAND9AI didn’t appear overnight. It came from years of shipping AI systems, testing ideas in public, and refining a constraint-driven, evaluation-first mindset.

Read expanded architecture notes (technical deep-dives) →

MD
Mustafa Darras
AI Systems Architect · LLM Evaluation, Alignment & Production Systems
The Constraint Mindset

Constraints in AI aren't the enemy. They're the catalyst.

We know the limits in machine learning: compute ceilings, token caps, model size vs. inference speed, deployment hurdles. These constraints force us beyond chasing “ideal” scenarios. We build what's real, efficient, focused, and deployable — and that’s where true breakthroughs happen.

Havana’s 1950s cars taught me this: innovation isn’t a buzzword, it’s a daily reality under pressure. That’s the same energy I bring to BAND9AI’s scoring engines and simulation flows.

#TheConstraintMindset #AIEngineering #LLMs #ConstraintDrivenDesign
AI Millionaire Model · JT Foxx

Turning AI into lead-generation and revenue systems, not demos.

The AI Millionaire Model certification reframed how I think about AI in business: it’s not about flashy capabilities, it’s about automating lead generation, compressing operations, and building systems that scale without burning out the operator.

BAND9AI borrows that exact playbook — product-led funnels, retention loops, and automated scoring/reporting pipelines — so the “AI” is never just a feature. It’s the engine behind a real business.

#AIMillionaireModel #BusinessAutomation #RevenueSystems
Before BAND9AI

InterviewAI, RLHF at Scale AI, and testing ideas in public.

Built & launched fast · MVPs under pressure

InterviewAI.ca was my first large-scale “simulation engine” — a voice-interactive mock interviewer that reads your resume, adapts questions, and scores answers in real time. It taught me how to combine NLP, evaluation logic, and UX pressure into something candidates actually use.

At Scale AI and Mindrift, I worked on RLHF and large-model evaluation: test prompts, pass/fail criteria, regression checks, drift monitoring. Those same ideas now power BAND9AI’s IELTS scoring pipelines.

From ChatGPT’s RLHF playbook to my own products, one theme holds: no eval, no product. Feedback loops are where systems become trustworthy.

#InterviewAI #RLHF #BuiltInPublic

From profile to product

You shouldn’t have to gamble
with a $300 exam.

BAND9AI exists so your IELTS journey feels like a well-played hand, not a desperate roll. Realistic simulations, evaluator-grade scoring, and feedback that actually moves your band.

Start a full mock test See pricing & plans