Chat - Bila UiTM Cuti
An AI chat assistant that answers UiTM academic calendar questions, providing accurate dates and information for students.
Live: bilauitmcuti.com/chat
Part of Bila UiTM Cuti — the companion chat (sometimes called UiTM Assistant / AI Chat UiTM) helps students check semester dates, lecture weeks, fee deferment windows, holidays, exams, and related UiTM info in a short conversation.
Students often need quick answers to questions like: when is semester break, what are the fee deferment dates, which dates correspond to weeks 1 to 14, or whether today is a holiday. Bila UiTM Cuti Chat provides these answers in a conversational format, based on the academic session and campus group chosen by the user.
The product stays focused on calendar and UiTM orientation questions. It is not a general-purpose chatbot for coursework, grades, or unofficial campus rumor.
Users can pick a Cloudflare Workers AI model in the chat composer. All picker models support function calling (tool use).
The chat composer lets users pick from multiple Cloudflare Workers AI models, including Gemma 4 (default), Llama 4 Scout, Mistral Small 3.1, and Nemotron 3 Super. All available models support function calling, enabling the assistant to look up academic facts before answering.
Function calling lets the assistant look up the right academic facts before answering. Reasoning-capable models can show a short thinking/reasoning section in the chat UI for harder questions.
| Feature | Description |
|---|---|
| Streaming chat | For academic calendar and UiTM general questions |
| Tool-calling agent | Supports function calling across all model pickers |
| Reasoning | Available for Gemma 4 and Nemotron 3 Super models |
| Session/program scope picker | Ensures answers match the user’s campus group and semester |
| Markdown replies | Supports tables, code, math, and diagrams where relevant |
| Cloudflare Turnstile | Provides bot protection on chat |
| Adaptive language | Shared language-control pipeline across every model (see below) |
Replies follow the same language behaviour on every picker model (Gemma 4, Llama 4 Scout, Mistral Small 3.1, Nemotron 3 Super) — switching models should not change how language is chosen.
- Match the user — English stays English; Bahasa Melayu Malaysia stays BM; natural Malay–English mix is mirrored when you code-switch.
- Loanwords are fine — Asking in English with UiTM terms like cuti or sesi still gets an English answer.
- Malaysian Malay, not Indonesian — BM replies use Malaysian wording and month names (e.g. Mac, Ogos), not Indonesian forms.
- Conversation memory — Short follow-ups (“Next”, “Okay”, “Lepas tu”) keep the thread’s language; a clear full-sentence switch (or “reply in English” / “jawab dalam BM”) updates the preference for that turn.
- Quiet quality check — If a reply is clearly the wrong language, the server regenerates once before you see it. Detection is never explained in the chat UI.
| Area | Choice |
|---|---|
| App | Next.js (App Router), React, TypeScript |
| UI | Tailwind CSS, shadcn/ui, Base UI, ai-elements |
| Runtime | Cloudflare Workers via @opennextjs/cloudflare |
| Inference | Cloudflare Workers AI + AI Gateway |
| Bot protection | Cloudflare Turnstile |
| Tests | Vitest |
- User submits a question via the chat interface.
- The system detects the user's preferred language (English, Malaysian Malay, or mixed) for this turn, automatically adapting if you switch during the conversation.
- The chat engine determines whether to use the tool-calling agent for in-depth answers or a simpler direct response.
- If the tool-calling agent is used, the model may access tools and generate a draft answer; models with reasoning skills can show their thought process in the UI.
- The reply is checked for accuracy—including language and calendar date correctness—before displaying to the user.
- Users can provide feedback with a thumbs up or down to help refine future responses.
- Bila UiTM Cuti — calendar
- Cloudflare Workers AI
This repository is public for transparency. Contributions and write access are limited to collaborators. See Contributing and the Code of Conduct.