STUDIO.01 — ABOUT

We engineer the layer around the model.

Recode Asia is an AI engineering studio for agents, retrieval, automation and operational software.

See the patternsSend a brief
STUDIO.02 — WHAT WE BUILD

Systems around real work.

Recode Asia designs AI agents, retrieval systems, automation pipelines and operational web software around the way a team already works.

The model is one component. The useful system also needs context, permissions, deterministic controls, evaluation and a clear route to a person.

FULL CAPABILITY LEDGER

Seven ways we build.

The architecture changes with the constraint. The engineering standard does not.

  1. 01

    AI agents

    Agents that read, decide and take bounded action across your inbox, CRM and internal tools — with approval gates and a trace for every run.

    tool use / approval gates / traces

  2. 02

    LLM harnesses

    Control layers around models: routing, structured outputs, fallbacks, caching, observability and evals.

    evals / routing / fallbacks

  3. 03

    Automation pipelines

    Deterministic workflows with queues, retries and alerts that move data and work between systems without manual handoffs.

    webhooks / queues / retries

  4. 04

    Second brains

    An org-knowledge product that makes documents, decisions and internal expertise searchable through one permission-aware interface.

    knowledge graph / search / access control

  5. 05

    Production-grade RAG

    Retrieval infrastructure that grounds answers in your source material, with deliberate chunking, citations and evals for recall and faithfulness.

    grounding / chunking / citations

  6. 06

    CRM/ERP web apps

    Operational software shaped around your workflows, roles and reporting — not a template your team has to work around.

    workflows / permissions / dashboards

  7. 07

    Conversion pages

    Fast, instrumented pages that explain the offer, remove friction and turn qualified attention into a measurable action.

    webgl / gsap / analytics

STUDIO.03 — APPROACH

Start with the decision.

We begin with the workflow, the evidence it depends on and the action that follows. Then we define where AI belongs, where deterministic code wins and which cases must stop for review.

A production plan follows from those boundaries: inputs, tools, evaluation cases, failure routes and the smallest useful release.

STUDIO.04 — OPERATING PRINCIPLE

Control stays visible.

Prompts, retrieval, tools, outputs and approvals are treated as parts of one inspectable system. That makes changes easier to test and uncertain cases easier to stop.

THESIS.01 — SIGNAL

The model can read the corpus. The work is pointing it at the right thing.

The model is not the strategy. We identify the decision, evidence and action that matter — then engineer the system around them.

Real-world applications

Where the leverage is.

Nine industries that already run on data — scattered across screens that don't talk. The revolution isn't new machines; it's the systems that finally coordinate the ones you have.

01

Ports & shipping

Berth windows, container dwell and discharge planning as one scheduler.

Dwell6d → 2–3d
02

Rail freight

Yard assembly, crew rotations and brake checks sequenced automatically.

Yard moves−30–45%
03

Warehousing

Pick paths re-slotted nightly from actual order flow.

Picks / hour+20–35%
04

Construction

Drone progress scans reconciled against BIM weekly; defects flagged, not discovered.

Rework−15–25%
05

Agriculture

NDVI scouting, irrigation scheduling and yield prediction per plot.

Yield+8–15%
06

Solar & wind

String-level thermal monitoring with maintenance ranked by output at risk.

Downtime−20–40%
07

Manufacturing

OEE watch with drift detection before the spindle fails.

Unplanned stops−25–50%
08

Logistics fleets

Dispatch, hours-of-service and deadhead optimization across the fleet.

Deadhead km−10–20%
09

Finance

Reconciliation, KYC refresh and exception queues as one continuous flow.

Manual recon−60–80%

Ranges from comparable builds; yours gets measured before any commitment — see the honest fit check on start a build.

Proof — delivery readouts
Brief → first working version10–15days
First version → production2–4weeks
Retrieval latency<200ms
Uptime across builds99.9–99.98%

Ranges, not points — every build is scoped against your systems before the clock starts.