Discuss your project
AI Solutions

Generative AI development and custom AI assistants

General chatbots know the world but not your organisation. We build AI assistants and applications that answer from your policies, procedures, products, contracts and records, cite sources, and respect who may see what.

Discuss your project How the work is delivered

A good fit if

  • Staff spend hours searching policies, procedures, manuals or past project files.
  • Information is spread across SharePoint, file shares, ERP, CRM and other systems.
  • You want an assistant in your app, portal or intranet, citing approved sources only.
  • You must control where data is processed, which model runs and what is logged.

What is included

  • Use case: users, questions, sources, permissions and success measures
  • Source review and ingestion that keeps documents up to date
  • RAG answers with citations, from content each user can access
  • Integration into intranet, Microsoft Teams, portal or business app, phone-friendly
  • Guardrails for off-topic use, personal data and prompt injection
  • Evaluation set and accuracy results; logging, usage and cost monitoring

What this service is

We design, build and support assistants and applications that answer from your own information, built on large language models. The model matters less than the work around it: sources, retrieval, permissions, accuracy testing and connecting to where people work. Typical examples:

  • Staff assistant for HR, finance and IT policy
  • Product assistant for sales and service
  • Contract clause finder
  • AI in your customer or partner portal

How it works

Most of our assistants use retrieval-augmented generation (RAG explained):

  1. AskSomeone asks a question where they already work.
  2. RetrieveApproved sources are searched with the user's own access rights.
  3. AnswerThe model answers only from those passages and cites them.
  4. CheckPeople open the source; if nothing is found, it says so.

Answers reflect your policies and products, not internet knowledge. For live data, such as an order status in Prometheus ERP or a CRM record, it calls those systems' APIs with the user's permissions; to take actions, we design an AI agent with approval points. Activity is logged per your retention rules. Model usage is metered and usually billed in US dollars, so we estimate and monitor running costs per question or user, including the effect of exchange-rate movements.

Platforms and data protection

We build on Microsoft Azure AI services, Amazon Bedrock, Google Gemini Enterprise Agent Platform (formerly Vertex AI) or open-weight models on servers you control, connected to Microsoft 365, SharePoint, file shares, databases, ERPs and CRMs, choosing enterprise services whose terms exclude your data from model training. Naming a product does not mean we partner with its vendor. There is no major public cloud region in Nigeria: large hosted AI services run abroad (Azure's and Google Cloud's nearest are in South Africa), and model availability varies by provider and region.

NDPA. The Nigeria Data Protection Act 2023 applies to personal data in an AI system, including the documents it searches; sending it to a model hosted abroad is a transfer outside Nigeria that needs a lawful basis. We choose services and hosting accordingly, check sector rules, minimise what an assistant can reach, document where data is processed and work with your data protection officer. General information, not legal advice; see AI governance and security.
Delivery

How the work is delivered

  1. Define

    Users, questions, approved sources, permissions model and success measures agreed.

    Output: Use case and evaluation plan.

  2. Prove

    A working prototype on representative documents, tested on real questions.

    Output: Prototype and accuracy results.

  3. Build

    All sources, permission-aware retrieval, guardrails and logging, integrated where people work.

    Output: Production-ready solution.

  4. Pilot

    Review a pilot group's answers and feedback; tune sources, prompts and retrieval.

    Output: Pilot report and improvements.

  5. Launch and monitor

    Roll out, train users, monitor accuracy, usage and costs against agreed thresholds.

    Output: Live assistant, runbook, monitoring.

Scope at a glance

Not included unless agreed

  • Model, platform and cloud usage, paid to the provider
  • Writing or restructuring source documents, unless agreed
  • Legal review of answers; post-handover tuning and support unless agreed

What we need from you

  • A business owner who decides when answers are good enough
  • Source access, sensitive data flagged, real questions with expected answers
  • Your data protection, security and data location requirements

Before handover

  • Meets the agreed accuracy threshold on the evaluation set before launch.
  • Permission, prompt injection and data leakage tests run and documented.
  • Admins add, update and remove sources unaided, with a guide and runbook.

What affects the cost

  • Number, size and variety of sources
  • Permissions model and how it is enforced
  • Accuracy target and evaluation set size
  • Where it runs, integration, dollar-priced model usage

No package prices: estimates in naira from an agreed scope, taxes shown separately.

Questions buyers usually ask

How is this different from ChatGPT or Microsoft 365 Copilot?

General assistants answer from their training and whatever you give them. Ours answers only from approved sources, with your permissions and citations, inside your systems. Many organisations use both.

Will the assistant make things up?

It can give confident wrong answers. We answer only from retrieved sources, cite them, test against real questions and say when nothing is found. People should still check answers that matter.

Which model do you use, and can it run in Nigeria?

We choose on accuracy for your questions, cost, data terms and where it can run: Azure, Amazon Bedrock, Google Cloud, or open-weight models (Llama, Qwen, DeepSeek, GLM, Mistral) on your own servers or in a Nigerian data centre.

How do you keep answers up to date?

Sources are re-indexed on a schedule or when documents change, retired documents are removed, and administrators see what is included and when it was last updated.

Have documents nobody can find answers in?

Tell us who needs answers, where information lives and how sensitive it is. We will arrange a conversation and suggest a proof on your content.

Scope an AI assistant