We still build the systems and surfaces. Now with AI wired into the dev loop.
We're not an agent-only shop. The plumbing, the marketing surfaces, and the search content that brings traffic — built in-house on the same .NET, Node, Postgres, MS SQL, AWS and Azure stack we've shipped on for eighteen years. What changed in 2026 is that AI tooling is in the dev loop, and most new products earn at least one agentic surface.
The same stack we've shipped on since 2008. With AI in the dev loop and, where it earns its keep, on the surface.
.NET 8/9, Node 20+, Postgres 16, MS SQL Server 2019/2022, on AWS, Azure, AliCloud and Huawei Cloud. We work where you already operate, not where a methodology deck wants you to.
Agentic CLIs — Claude Code, Codex CLI, Gemini CLI — paired with a senior engineer in the dev loop. Faster scaffolding, faster review, fewer regressions. The product still ships on tested code, in a tested pipeline, with named humans on the merge button.
Most new builds now earn at least one narrow-agent surface — copilots, summarisers, structured extraction, classification. We design the safety envelope and the eval harness before the agent ships to a user, not after the first incident.
Brand-grade marketing sites and product surfaces, on infra we operate.
Design system and tokens, site implementation, headless CMS or MDX-driven content, deploy infra. Astro for marketing surfaces, Next.js for app-shaped ones. Typography and layout grounded in a living style guide, not a one-off Figma file.
Snapshotted releases, atomic rsync deploys, rollback by release ID. We don't hand the keys to a third-party platform we can't reach at two in the morning. If your CDN edge sneezes, we have a shell and a runbook, not a support ticket.
This site is our own exhibit. The pattern you see — typography, taxonomy, agent-team deep-dives, field notes, the deploy and rollback discipline — is what the engagement looks like when it lands on your domain.
Volume without losing voice. A research desk, not a generator.
- Article Studio is the engine. Nine stages, ten specialists, a tier-graded source registry. Every claim cited, every URL fetched live, bilingual EN ⇌ 简中 where needed. Documented end-to-end at /agent-teams/article-crew/.
- Singapore and APAC, primary-source. SFA, IMDA, ESG, A*STAR, MOH and provincial-CN primaries are read in their own language and tracked back to source. Local angle is not a synonym for "we asked an LLM about Singapore."
- Search and generative-engine optimisation. The piece is structured to be retrieved and quoted by both — schema where it earns its keep, semantic structure that survives a chunker, citation density that an answer engine can lean on without hallucinating.
- Senior editor, non-skippable. The editor reads every piece end-to-end before publish. Listicles without direct project experience are refused. "What is [buzzword]?" explainers without an opinion are refused. Voice over volume when the two collide.
- Engagement shape. Either as a content retainer (monthly throughput against an editorial calendar) or as a one-off pillar build (a clutch of cornerstone pieces around a launch). We will tell you which during the diagnostic.
What people ask before they brief us.
- What stack do you build on?
- .NET 8/9, Node 20+, Postgres 16, MS SQL Server 2019/2022, on AWS, Azure, AliCloud and Huawei Cloud. We work where you already operate, not where a methodology deck wants you to. Astro for marketing surfaces, Next.js for app-shaped ones.
- Do you also write SEO and GEO content?
- Yes — through the Article Studio crew. Nine stages, ten specialists, a tier-graded source registry, every claim cited and every URL fetched live. Bilingual EN ⇌ 简中 where needed. Either as a content retainer against an editorial calendar, or as a one-off pillar build around a launch.
- Do you operate the infrastructure after launch?
- Yes. Snapshotted releases, atomic rsync deploys, rollback by release ID. We don't hand the keys to a third-party platform we can't reach at two in the morning. If your CDN edge sneezes, we have a shell and a runbook, not a support ticket.
- Will you bolt on agentic surfaces, or is that a separate engagement?
- Same engagement. Most new builds now earn at least one narrow-agent surface — copilots, summarisers, structured extraction, classification. We design the safety envelope and the eval harness before the agent ships to a user, not after the first incident.
A paid one-week diagnostic up front decides scope, stack and engagement shape. You walk away with a written proposal whether or not we go further.
Brief us →