We Don’t Need Bigger AI. We Need AI Infrastructure That Pays Its Way.
From an accounting and asset-management standpoint, the first major foundation is a matured Detached System running on a LAMP-based architecture. The principle is straightforward capital allocation — not every task needs an LLM. Validation, business rules, reconciliation, workflow control, repetitive logic and many operational decisions can be handled outside the model, preserving cash for decisions that genuinely require intelligence.
In the last six months, while the overall platform is still only partially completed, we have already built approximately 250 Detached Systems. The total development cost has been less than USD200, including experimentation, failed approaches, repeated testing and many mistakes along the way. In our books, that is a unit cost that makes commercial sense for Malaysian SMEs, proving that useful business systems do not always require heavy capital expenditure.
The second layer now in progress is our modded AI agent with Smart Routing capability. Instead of automatically routing every task to the largest model available, the system is being designed to continuously identify the smallest and lowest-cost LLM capable of completing each task reliably. This is essentially dynamic cost optimisation: matching the right asset to the right job.
The logic is straightforward: simple task → small model, difficult task → stronger model, no intelligence required → Detached System. The objective is not merely to reduce token usage, but to make AI infrastructure economically sustainable by allocating compute and model spend only where intelligence is genuinely required.
The third layer is our own specialised LLM models, designed specifically to work together with our Detached Systems. We are currently experimenting with 7 open-source LLM models as the foundation for this work. Hugging Face will be part of our technology ecosystem, supporting access to the open-source model ecosystem, datasets, training tools and infrastructure. For an SME budget, open-source foundations reduce licensing drag and keep the balance sheet lighter.
But these three developments are ultimately aimed at something much bigger: lowering the total cost of ownership for business systems so that ordinary SME owners can participate, especially businesses that do not have unlimited funding, GPU capacity, IT teams or technical resources. It is a financing and access question as much as a technology one.
One day, even a makcik selling pisang goreng by the roadside should be able to digitise her operations without hiring an IT department. She should simply be able to say, “Manage my stock, calculate my daily profit, monitor ingredient costs, remember my regular customers and tell me when I need to buy more bananas,” and the agent should help assemble the system behind it. For us in finance, that means turning everyday business decisions into trackable, auditable records.
Our direction is clear: Detached System → Smart Routing → Lowest Suitable LLM → Specialised Own LLM → AI-Assisted System Development for Everyone. Minimum cost is the immediate objective. Free is the dream. We are not trying to build the biggest AI; we are trying to make AI and system development small enough, affordable enough and simple enough that even the smallest SME can justify it on the P&L and build with it.
#AINNA #NeuralOps #HuggingFace #OpenSourceAI #LLM #AIInfrastructure #SmartRouting #DetachedSystem #SME #SystemDevelopment #Automation
