Production AI for Nigerian businesses — not demos.
We design and ship AI features that hold up in production: RAG-powered search, document automation, voice and chat assistants, intelligent workflows, and AI agents — built on Anthropic Claude, OpenAI, and open-source models, with real evals and observability.
What we build
Retrieval-augmented generation (RAG) systems
Smart search over your documents — policies, legislation, contracts, knowledge bases — that returns cited, grounded answers. Used in our Rivers State House of Assembly platform for citizen-facing legislative search.
AI assistants & chatbots
Embedded AI assistants that answer questions grounded in your business data — used in Every27 for accounting Q&A and Growzen for seller support.
Document automation
Extracting structured data from invoices, contracts, ID cards, and forms — at production scale, with explicit confidence scoring and human-in-the-loop review for edge cases.
AI agents & workflows
Multi-step agents that take actions: triage tickets, schedule appointments, route customer enquiries, draft responses for human review. Built on Claude and OpenAI tool-use APIs with proper guardrails.
Voice & transcription
Speech-to-text, voice assistants, call summarisation — including Nigerian-accented English and pidgin handling.
Custom model fine-tuning
When prompting isn't enough, we fine-tune open-source models (Llama, Mistral, Qwen) on your data — and host them where you control the infrastructure and the privacy posture.
How we ship AI to production
- Eval-first. We define success metrics before we write a prompt. Every change is graded against a held-out eval set.
- Observability. Every AI call is logged, traced, and replayable — no mystery hallucinations.
- Cost discipline. Prompt caching, tiered model routing (Haiku / Sonnet / Opus), and request budgets so you know what AI is costing per user.
- Safety & data privacy. Customer data never trains a public model. NDPR-aligned data handling.
- Graceful degradation. If the model is slow or down, the product still works.
FAQ
What's the cost of an AI feature?
A scoped AI feature (e.g. "AI search over our document library") starts from ₦3M build + monthly LLM costs (typically ₦50k–₦500k depending on volume). Full AI products are quoted separately.
Which models do you use?
We default to Anthropic Claude (best reasoning, strong instruction-following) and OpenAI GPT-4 / 4o. For voice, we use ElevenLabs and Whisper. For self-hosted, Llama 3 / Qwen / Mistral.
Will my data leak to the model providers?
No. We use enterprise tiers with no-training agreements (Anthropic and OpenAI both offer this), and for sensitive workloads we deploy open-source models on infrastructure you control.