A Reference Point Built
for Deployment Decisions
draviak exists to give engineering teams a structured, neutral place to reason about hardware choices for neural network deployment.
← Back to HomeHow draviak Came Together
draviak started from a straightforward observation: engineering teams preparing to deploy neural network models were spending disproportionate time on hardware questions that had already been worked through elsewhere — by other teams, in other organisations — but the knowledge wasn't organised anywhere accessible.
The team that built draviak had worked across a range of deployment environments in Southeast Asia — from early-stage ML infrastructure at technology firms in Kuala Lumpur to more involved engagements at larger organisations — and recognised a consistent gap between what teams needed to know and what was readily available to them.
The response was a structured information resource built around the actual shape of the deployment decision: what model characteristics drive hardware requirements, how those translate into infrastructure choices, and where the common tradeoffs sit. That resource became the foundation for everything draviak offers today.
Our Mission
To organise deployment hardware knowledge into clear, usable routes — so engineering teams spend less time searching and more time building.
Our Position
We hold no commercial relationships with hardware vendors. Our guidance reflects the landscape as we understand it, written to inform rather than direct.
Our Approach
Every offering is structured around the deployment decision itself — the model's requirements, the serving environment, and the operational constraints that matter in practice.
People Behind the Atlas
Nadia Farhan
Nadia led ML infrastructure planning at a Kuala Lumpur-based fintech before joining draviak. She structures the library content and oversees engagement quality.
Reza Hakim
Reza focuses on hardware landscape research and keeps the guidance library current with quarterly update cycles. His background spans systems engineering and inference optimisation.
Siti Wulandari
Siti coordinates planning workshops and custom engagements, ensuring each team's questions are properly scoped before any work begins. She holds a background in technical programme management.
Standards We Hold Ourselves To
Vendor Independence
No commercial relationships with hardware vendors means the guidance reflects our honest assessment of the landscape, not a preferred outcome.
Regular Content Review
The deployment hardware landscape evolves. Library content is reviewed and updated on a quarterly cycle, with material changes flagged to subscribers.
Client Data Privacy
Information shared during engagements is used only for the purpose of that engagement. We do not share organisation-specific details with third parties.
Scoped Engagements
Each custom engagement begins with a discovery phase to ensure the guidance delivered actually addresses the team's stated questions rather than a generic brief.
Written Deliverables
Every engagement produces a written output — not just a verbal debrief. Teams should be able to revisit the guidance as their deployment context changes.
Direct Communication
Teams work directly with the people producing the guidance — no account management layers between the question and the person who can answer it.
Deployment Hardware Knowledge for Neural Network Workloads
Neural network deployment involves a particular set of hardware considerations that differ meaningfully from traditional software infrastructure. The characteristics of a model — its architecture, parameter count, precision requirements, and expected query volume — each place different demands on the hardware that serves it. draviak's guidance organises these relationships so engineering teams can reason about them systematically rather than by trial and error.
The Malaysian technology sector has seen a significant expansion in ML workloads over recent years, with engineering teams across fintech, e-commerce, and enterprise software working to move models from development environments into production. Hardware selection often becomes a bottleneck in this process — not because the information doesn't exist, but because it exists across scattered sources that weren't written with the deployment decision in mind.
draviak addresses this by structuring information around the decision itself. Reference articles in the Guidance Library cover how to characterise a model's hardware footprint, how to compare CPU, GPU, and purpose-built inference hardware against workload requirements, and how to think about the tradeoffs between throughput, latency, and cost at different serving scales. The Planning Workshop builds on this foundation with a facilitated session that applies the framework to a team's specific workload. The Custom Engagement goes further, producing a written reference tailored to an organisation's deployment questions.
draviak operates out of Kuala Lumpur and works primarily with engineering teams in Malaysia and the broader Southeast Asian region. Engagements are conducted in English.
Find the Right Route for Your Team
Whether you're researching the landscape or ready to work through a specific deployment question, we're here to help you navigate it.
Get in Touch