NAME
danielsam - Daniel Shamsudin, software engineer, founder of CustomerMaestro
SYNOPSIS
danielsam [--build product] [--ship to-production] [--self-host] [--research question]
DESCRIPTION
I build and run software for small businesses and schools in Malaysia. I write the code, deploy it, keep it secure and answer the support email. Most of what I make runs on servers I maintain myself.
I care about systems that stay correct when nobody is watching: money stored in integer sen, marks computed by code instead of a language model, and every number on a page traceable to its source.
CUSTOMERMAESTRO
I founded CustomerMaestro to build practical software for Malaysian businesses. It runs three products.
- CariRumah homestay booking
-
A booking platform for homestay hosts and guests, with listings, availability, payments through CHIP and host payouts. I also run its security reviews: a full audit and an authorized test of the live site, with every finding fixed and deployed.
Next.js, Clerk, TypeORM, PostgreSQL, MinIO, CHIP
- Maestro Invoice invoicing and accounts
-
Invoices, bills, expenses and payments for small companies, ready for LHDN e-Invoice: tax types per line, buyer and supplier identity fields, and role-based access with an audit log. Reports include profit and loss on accrual or cash basis, expenses by category, receivable and payable aging, and a tax summary.
Next.js, Prisma, PostgreSQL, NextAuth
- LetsMaths exam practice for primary school
-
A maths practice app for Malaysian pupils, built with my co-founder Shafiq. I built the question engine: it generates bilingual questions and full exam papers aligned to the KPM DSKP syllabus, and it marks every answer deterministically. A language model only explains; it never sets a number or decides a mark.
Python, FastAPI, Next.js, PostgreSQL
RESEARCH
- Sentiment Parliament Dewan Rakyat Hansard
-
Profiles Malaysian MPs from what they say in official parliamentary debate, for citizen transparency. Each speaker turn is parsed from the Hansard PDFs and extracted into its targets, stance, rhetorical mode and tone, instead of being scored with a bag of words that fails on negation, sarcasm and code-switching between Malay and English.
On 2,102 turns from 9 sittings, unsupervised clustering separated the government front bench from the rest of the chamber using speech alone. The pipeline was never told who is a minister or which party anyone belongs to.
The work reports its limits openly: weak cluster structure, thin evidence for half the chamber, and a stance metric that turned out to measure the matcher rather than the model. Party is never inferred from speech, and the scope stays on official conduct.
Python, LLM extraction with provenance, k-means
INFRASTRUCTURE
The products above run on hardware and servers I set up and maintain: Docker services behind Cloudflare Tunnel and Caddy, deploys that go live when CI passes and roll back when a health check fails, fail2ban jails for exposed services, and backups before every database migration. Part of it runs on a 2011 dual-core desktop.
OPEN SOURCE
- wmux
- A small tmux-style terminal multiplexer for Windows, built on ConPTY. Rust.
- PowerLight
- A Spotlight-style search launcher for Windows. Rust.
- pacmap
- Graphs explicitly installed pacman and AUR packages to find the dormant ones taking disk space. Python, with HTML and GTK4 front ends.
- seriousdb
- A small persistent key-value store over HTTP. Python.
SEE ALSO
- danielshamsudin8@gmail.com
- customermaestro
- admin@customermaestro.com, customermaestro.com
- github
- github.com/danielshamsudin
- linkedin.com/in/danielshamsudin