Services & pricing
Singapore-based bilingual developer who ships production AI systems.
Four fixed-scope packages with prices on the page, so you know what you are buying and roughly what it costs before we talk. Not sure which one fits? Book a free 15-minute call.
Packages
What you get, and what it costs
WhatsApp / website AI receptionist
From S$1,500+ S$100/month maintenance
For: Restaurants, clinics, salons and service businesses answering the same questions all day.
- Answers customers 24/7 in English and Chinese, trained on your menu, prices and policies
- Takes bookings and orders and writes them into your booking or order system
- Hands the chat to a human when it is unsure, with the conversation so far
- Monthly fee covers hosting, monitoring and small content updates
Ask about this package on WhatsApp →Invoice, receipt & ordering automation
S$2,000 – 5,000
For: Teams retyping supplier invoices, delivery orders or WhatsApp orders into spreadsheets by hand.
- Photos, PDFs and chat messages read with OCR + LLM extraction
- Clean rows pushed into Google Sheets, Xero or your ERP
- Low-confidence fields and mismatched totals flagged for a human to check
- Price depends on the number of document types and the target system
Ask about this package on WhatsApp →Internal knowledge base / SOP bot
S$2,000 – 4,000
For: Multi-outlet businesses where new staff keep asking the same SOP questions.
- Your SOPs, manuals and policies (Word, PDF, Notion) turned into a searchable knowledge base
- Staff ask in English or Chinese on WhatsApp, Telegram or a web page
- Every answer cites the source document and section
- Update it by replacing a document, no retraining needed
Ask about this package on WhatsApp →AI-written code rescue
US$1,500 – 8,000
For: Apps built with Cursor, Claude Code, Lovable or Bolt that work in the demo but are not ready for real users.
- Security: leaked keys, broken auth, injection and data exposure
- Deployment: domain, HTTPS, backups, CI and one-command rollback
- Performance: slow queries, N+1 calls, caching and cost of model calls
- A written report plus the fixes; price depends on codebase size
Ask about this package on WhatsApp →- Fixed quote in writing after a free 15-minute call; no hourly billing surprises.
- Third-party costs (WhatsApp Business API, model usage, hosting beyond the basics) are passed through at cost.
- 0115-min callFree. We map the problem and pick a package.
- 02Fixed quoteScope, price and timeline in writing.
- 03Build & demoYou see working software early and often.
- 04Launch & handoverDeployed, documented and monitored.
Case studies
Systems already running in production
F&B operations SaaS
MakanCloud
- Problem
- Multi-outlet F&B brands in Singapore ran outlet ordering over WhatsApp and paper slips, central-kitchen production was hard to trace, and rosters were built by hand every month.
- What I built
- One platform covering outlet ordering, central kitchen, delivery dispatch, OR-Tools rostering, offline-capable POS and QR ordering, multi-tenant and white-label.
- Result
- Live and sold commercially at makancloud.com, from S$59/month
- Designed, built and operated by one person with AI agents
See MakanCloud →Booking + review workflow
Creator & mystery-shopper booking
- Problem
- A restaurant brand coordinated influencer visits and mystery-shopper checks over chat and spreadsheets: slots got double-booked, reports and receipts were scattered, and bank details for reimbursements sat in chat history.
- What I built
- A self-serve booking link per brand with per-outlet calendars; side-by-side compare and approve; mystery-shopper invite links with an on-site questionnaire, photos and receipts; review and payout flow with a regional-manager role.
- Result
- Live at book.makancloud.com, bilingual, multi-brand and multi-outlet
- Bank details AES-256-GCM encrypted, masked by default, every reveal audited and auto-purged after retention
- Excel export and end-to-end tests covering the full booking and payout flows
Visit the live system →Autonomous AI research pipeline
Daily investment briefing
- Problem
- Following AI, robotics, space and biotech names across US, Hong Kong, China A-share and crypto markets took hours of reading every morning.
- What I built
- A pipeline that runs at 06:00 SGT: prefetches market data, writes the briefing with a multi-model fallback chain, passes it through a quality gate, publishes it with an English summary, and reviews itself every week and month to tune its own prompt.
- Result
- 138 issues published, no manual steps
- 81 names tracked in a rule-based model portfolio with entry and exit triggers
- Retries, fallbacks and failure alerts keep it running unattended
Read the briefings →Not sure which one fits? Start with 15 minutes.
Tell me about your business and where it hurts. I will tell you which package fits and roughly what it costs — or say so if none does.