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 global goal: predictable IELTS outcomes.
Most students treat IELTS like a casino, pay, hope, and wait. BAND9AI rolls the dice in advance: full mock tests, examiner-style scoring, and pressure-based simulations so test day feels familiar, not random.
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.
Founder
Mustafa Darras
AI Systems Architect · BAND9AI Human Systems Inc.
A deck built for outcomes
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
I design multi-module AI systems. Listening, Reading, Writing, Speaking, that behave like a real examiner: timed pressure, rubric-aligned scoring, and unified logic across skills.
Joker Removed
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.
Full Deck in Play
I operate with a growth-first mindset shipping fast, validating distribution early, and turning AI into revenue, not just demos.
Constraint mindset
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
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.
Mustafa Darras
AI Systems Architect · LLM Evaluation, Alignment & Production Systems
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.
AI Millionaire Model · JT Foxx
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.
JT Foxx · MEGA Events
A strategic framework that blends AI with business execution: automated lead engines, scalable offers, and systems that sell while you build.
Before BAND9AI
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.
From profile to product
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.