Proposal
Carbon GPT
CGPT Consulting Sdn Bhd
BLK C, Level 17-08, Jalan 16/20
46350 Petaling Jaya, Selangor
+60 17 2356 777 · carbongpt.ai
5 May 2026
Prepared for
Monash University Malaysia, Finance Helpdesk

Proposal for AI Finance Helpdesk · Phase 1

Video Tour of This Admin Portal

A short walkthrough of this working proof-of-concept chat, escalation, and analytics.

The Problem

Your Finance Helpdesk receives the same routine questions every term. The answers already exist in your website, but students rarely find them.

The cost is twofold:

  1. Students with genuine, time-sensitive issues (payment plans, encumbrances blocking exams) wait behind students who could have self-served.
  2. Your team is consumed by volume especially during enrollment period.

The Solution

An AI assistant grounded in your approved sources. It answers routine questions with citations, and escalates the rest with a drafted reply your officer can edit and send.

Students get plain-English answers in a chat widget on your finance pages.

Officers get a triage queue: conversation, draft reply, suggested actions, and policy links. Review and approve rather than write from scratch.

Scope of Phase 1

Student chat

  • Embeddable widget on monash.edu.my, with session memory and 1–5 rating.
  • Answers grounded in approved sources, with clickable citations.
  • Fee calculator logic coded into the pipeline.
  • Escalation when staff judgment is needed.

Knowledge base

  • Seeded with all SOW Part 1 sources.
  • Upload tool for internal policies, FAQ exports, and historical tickets.

Admin triage queue

  • All escalations sortable by priority, status, sentiment, category.
  • AI-drafted reply and 3–5 suggested next actions per ticket.
  • One-click templates, internal notes, sentiment flagging.

Reporting

Dashboard with sessions, deflection, escalation, satisfaction, sentiment, and top categories. Exportable for finance committee reporting.

Technical foundation

  • Frontend: Svelte (Typescript)
  • Backend: Hono.js (Typescript)
  • AI model: Google Gemini 2.5 Pro by default. We can swap in any current frontier thinking model from OpenAI, Anthropic (Claude), or Google at no change to the build fee.
  • Database: Neon Postgres (managed, point-in-time recovery)
  • RAG (vector database): Qdrant
  • Hosting: Vercel

Not in Phase 1 (deferred to Phase 2, quoted separately)

  • SMS / Student Information System integration. Account-specific questions escalate to your team in Phase 1.
  • Asana sync (two-way) with your existing finance helpdesk project.
  • Outbound email send and inbound email threading back onto tickets.
  • Monash SSO. Phase 1 admin access is via email OTP.
  • Mobile push notifications for your team.
  • Multi-officer routing, per-ticket SLA timers and breach alerts, and PDPA data export/deletion workflows.
  • Bahasa Melayu and voice/audio input.

Commercial Terms

1. Build fee: RM 40,000 (flat, one-time)

Split 50/50: deposit on signing, balance on go-live. Net 15 on each invoice. Subscription invoiced monthly in advance. All amounts exclusive of SST.

2. Subscription (starts on go-live)

Standard
RM 2,500 / month
or RM 30,000 / year

Includes LLM token budget for 3,000 enquiries/month.

Bring-Your-Own AI API key
RM 1,800 / month
or RM 21,600 / year

You provide a Google API key and pay Google directly for AI usage.

Standard tier includes:

  • Hosting, infrastructure, and security patching.
  • LLM token budget for 3,000 enquiries/month (one enquiry = one chat session). Overage at RM 1.20 each, with alerts at 80% of cap.
  • Mon–Fri 9–6 MYT support, 1-business-day response SLA, 99% uptime target.
  • KB updates within 1 business day.
  • Up to 4 hours/month of AI tuning and calibration (fixing wrong answers, prompt tweaks, retrieval adjustments). Unused hours don't roll over; beyond 4 hours, billed at RM 300/hour. Major changes (new categories, new policy logic, restructuring) are quoted as separate work.
For comparison: Intercom Advanced + Fin AI for an equivalent setup runs roughly RM 8,000–16,000/month (USD 99/officer + USD 0.99 per AI resolution, depending on seat count). That's 3–6× the cost of this proposal at the same monthly volume.

3. Term & exit

Month-to-month from go-live. Cancel anytime with 30 days' written notice. Any prepaid months are refunded pro-rata. No lock-in, no minimum term.

Phase 2 features are quoted as separate engagements and are not included in the subscription.


After you sign

  • Deposit invoice issued.
  • We seed your KB and provision infrastructure.
  • We fine-tune the RAG and AI on your sources and historical tickets.
  • Production hardening, UAT, and security sign-off.
  • Go-live 2 months from signing. One month post-launch warranty and tuning window included.
  • Subscription begins. Cancel anytime with 30 days' notice.

What we need from you

  • Approved sources and any historical material to index.
  • Gated sources: Fee statement in USD, Payment extension, Uplift encumbrance, Student Fees Refund Procedure PDF, and any internal "to be provided" material. Several SOW links 404 from outside the Monash network.
  • FAQ database in any format (CSV easiest). Optional but valuable: 200–500 anonymised past tickets to calibrate tone.
  • Embed targets (default monash.edu.my/finance/*), allowlisted origins, and an IT/security contact for CSP review.
  • A nominated finance team owner to review AI-drafted replies and flag edge cases for calibration.
  • Compliance: any PDPA / DPA template or infosec review form we should complete.
  • Logistics: calendar contact for kickoff and weekly 15-minute check-ins.

Acceptance

For Monash University Malaysia
Name
Title
Date
Signature
For CGPT Consulting Sdn Bhd
Name
Shawn Lim
Title
CTO / Co-founder
Date
5 May 2026
Signature
CGPT Consulting Sdn Bhd · carbongpt.ai