Most SA businesses don't need custom AI. They need a chatbot, an automation, a document processor — services that already exist as productised offerings. Custom is for when those don't fit: a workflow no off-the-shelf tool models, an integration too deep for standard connectors, a compliance requirement that rules out commercial APIs, or volume that demands ownership of the model.
When custom is the right answer, half-measures cost more than doing it properly. We build full-stack AI systems: data pipelines, retrieval infrastructure, agent orchestration, custom UIs, fine-tuned or self-hosted models — whatever the problem actually requires.
What we build
Fine-tuned models
Open-weight models trained on your data — Llama, Mistral, Qwen — for domain language, output formats, or cost-sensitive volume.
Multi-agent systems
Specialist AI agents collaborating on complex tasks — research, business processes, adaptive workflows.
RAG infrastructure
Retrieval-augmented generation grounded in your documents, databases, and live systems. Production-grade vector search.
Deep integrations
Tight integration with bespoke or legacy systems — ERPs, custom databases, on-premise infrastructure, mainframes.
Custom UIs
Bespoke applications where AI is embedded in a custom workflow — not an add-on widget.
Self-hosted deployment
Full control over model and data. ZA-region or fully on-premises infrastructure. No external API dependencies.
When custom development is right
Genuinely unique workflow. Your business does something the rest of the market doesn't, and standard automation tools force you to compromise. The compromise costs more than custom would.
Deep integration requirements. Your AI needs to talk to a 1990s mainframe, a proprietary CRM with no public API, or three industry systems strung together with custom logic. Off-the-shelf tools won't reach.
Data residency or compliance. POPIA cross-border restrictions, sector-specific regulation, or contractual obligations require ZA-only or fully-private infrastructure with no commercial API calls.
High volume, cost-sensitive. Once you're processing tens of millions of AI interactions, owning the model becomes cheaper than per-call API pricing. Self-hosted open-weight models pay back at scale.
Strategic differentiation. Your AI is part of your product or competitive moat — not a productivity tool. The capability has to live with you, not at a vendor.
How we work
Discovery (1-3 weeks). Workshops, system audits, data review, success-criteria definition. End deliverable: scoped technical specification with realistic effort estimate.
Build (8-24 weeks). Iterative delivery, weekly demos, working software in your environment from week 4. Production-deployment criteria defined upfront.
Production rollout. Phased traffic, monitoring, fallbacks, runbooks. Your team trained on operation.
Optional ongoing support. Retainer-based for monitoring, updates, model improvements. Or full handover with our help to internalise.
What it costs
Engagement size. R60,000 to R600,000+ depending on scope. Most custom projects land R120,000-R280,000. Anything under R60,000 we'd typically push toward our standard productised services where one of them fits.
Discovery alone. R20,000-R45,000 for a structured discovery and proposal — no commitment to build with us afterwards.
Code ownership. All bespoke code transfers to you on delivery, with documentation. No vendor lock-in.
How to start
30-min call to discuss what you're trying to do. We'll be honest about whether custom development is the right answer or whether something simpler would work. If custom is right, we'll outline a discovery engagement.
Email info@faautosolutions.com or use the contact form.