AI Tools & Products
AI tools and product development
AI earns its place by removing real repetitive work, not by demonstrating technology. We build AI systems that run day to day: connected to your data, embedded in your process, usable the day they are handed over.
What kind of AI tools do we build?
Practical tools grounded in a company's existing data: knowledge-base Q&A, customer-service agents, document extraction, content generation and process automation.
- RAG knowledge bases
- Turn your website, documents and product data into a retrievable knowledge base, so the AI answers from your actual material rather than generic training data.
- AI agent deployment
- Connect to WhatsApp, your website or internal systems to handle enquiries, bookings and initial screening — and hand over to a human when conditions are not met.
- Document processing
- Extract fields from invoices, contracts and reports, returning structured data your systems can use.
- Content generation tools
- Image, deck and product-marketing generators that output usable assets within your brand rules.
- Process automation
- Embed AI judgement into existing workflows for classification, summarisation and first-pass responses.
- Metering and cost control
- Token usage tracking and tiered pricing structures, so the cost of AI features stays predictable and attributable.
Case
Case: MeetBiz
A WhatsApp AI support platform for Hong Kong SMEs. A merchant submits their company URL, the system builds a RAG knowledge base automatically, and a dedicated AI agent is deployed within minutes.
Core mechanism
- Crawls the company URL and builds the knowledge base automatically, with no manual data preparation
- AI agent connects to WhatsApp and answers customer enquiries around the clock
- Scanning a QR code completes deployment, removing the technical setup step
Platform features
- AI image, deck and product-marketing generators
- Image and video processing, full PDF toolset
- Conversational search and two-way supply-demand matching
Commercial architecture
- Token metering, so usage maps directly to cost
- Tiers from a free trial allowance to unlimited enterprise plans
- Designed for SME users without a technical background
How does an engagement work?
Four stages, each ending at a clear decision point where you can choose whether to continue.
- 01
Define the problem
Establish the specific problem and how it is handled today, quantifying the time and headcount it currently consumes. If the assessment says it is not worth building, we say so.
- 02
Prove feasibility
Build a small prototype on real data and test whether accuracy reaches a usable level. The point of this stage is to find dead ends early.
- 03
Build and integrate
Develop the production version and connect it to existing systems, including permissions, logging and exception handling.
- 04
Launch and operate
Tune against real usage, refining the knowledge base and response rules from actual queries.
Frequently asked questions
- We are a small company. Is AI worth adopting?
- Size is not the deciding factor; repetitive volume is. If a task repeats dozens of times a day, follows reasonably clear rules, and currently occupies staff time, it is usually worth examining. We quantify the time actually saved before recommending a build.
- What is RAG, and how is it different from just using ChatGPT?
- RAG is retrieval-augmented generation: the AI retrieves relevant material from your own data before answering. A general-purpose AI can only draw on its training data and cannot answer accurately about your products, pricing or processes. RAG supplies that missing layer.
- How long does an AI tool take to build?
- It depends on scope. Connecting existing material to a knowledge base and launching a support AI is usually a matter of weeks; projects involving multiple system integrations or custom workflows take longer. We provide a phased timeline after assessment and ship a usable minimum version first.
- Will the AI give wrong or invented answers?
- It can — that is inherent to language models. We manage it by constraining the source material, requiring citations, defining conditions that hand the conversation to a human, and testing against real questions before launch. The goal is not zero errors but errors that are contained and traceable.
- How is data security handled?
- The architecture follows the sensitivity of the data. General business material can use commercial APIs; where personal data or trade secrets are involved we evaluate self-hosted models or data redaction. These arrangements are defined before the project starts.
Have a project in mind?
Tell us where the business is and where you want it to go. We assess feasibility first, then propose.