AI Engineering · Generative AI
A knowledge assistant that lets employees ask questions about company policies and processes in plain language — and see exactly which document every answer came from.
Runs in your browser with fictional sample data.
01The business problem
Company knowledge is spread across handbooks, policy PDFs, SOPs and playbooks. People ask colleagues the same questions every week, new joiners take months to find their way around, and generic chatbots give confident answers that aren't grounded in the company's own rules.
02The proposed solution
A retrieval-augmented assistant: documents are split into passages and indexed; each question retrieves the most relevant passages, and the answer is written only from them, with numbered citations back to the source. Questions outside the knowledge base are declined rather than guessed.
Sample organisation in the demo: Corvane Logistics (fictional).
03Key features
Natural-language questions, streamed answers and follow-ups that keep the context of the conversation.
Every claim links to the passage it came from, so people can verify before they act.
Search the knowledge base directly, with matching passages highlighted in each document.
Past conversations are kept in a sidebar and can be reopened or cleared.
Out-of-scope questions are declined, and loading and error states are designed, not left to chance.
Starter prompts help people discover what the assistant knows.
04Technical approach
The production stack we would propose. The demo itself is built with Next.js, React and TypeScript and runs entirely in the browser.
The UI talks to a single KnowledgeProvider interface (ask, search, documents). The demo uses an in-browser mock; production swaps in a fetch() to a retrieval API with no component changes.
Keyword-weighted retrieval runs over a prepared sample knowledge base. Answers are composed only from retrieved passages, the same constraint a production system enforces on the model.
Document ingestion and chunking, embeddings in a vector index, a retrieval service with re-ranking, an LLM constrained to cited context, and an evaluation set that runs on every change.
The demo is a working front end over prepared sample data. It is not connected to a live AI model or a production backend, and it doesn't collect any information.
Launch live demoKriyexa Intelligence Hub is a concept project built by Kriyexa to demonstrate capability, not a client engagement.
More work
Workflow Automation · AI Agents
02A visual automation builder for business workflows: an AI agent qualifies every new lead, routes it down the right branch and keeps the CRM up to date — with every step logged.
React · TypeScript · Workflow engine
View projectBusiness Intelligence · Analytics
03An analytics dashboard that brings revenue, sales, acquisition and conversion into one view — with plain-language insights generated from whatever period you select.
Next.js · TypeScript · SVG charting
View projectFull-stack Software · E-commerce
04A storefront and the operations console behind it: browse and buy from a product catalogue, then manage the resulting orders, stock levels and sales from the admin side.
Next.js · React · TypeScript
View projectWe'll adapt the approach to your data, your systems and the outcome you're after.