AI workflows and custom systems

We embed in your process,then build what streamlines it.

Most software gets specified in a meeting room and fails there. We start inside the operation instead, mapping how the work really flows and where it stalls, then build the systems and agents that clear it.

Small senior team. The engineer who mapped your process leads the build.

Start a conversationSee the work

What we do

Process mapping

We sit with the work before we touch a keyboard, mapping where it stalls and who is keying the same thing twice.

What the map finds decides which of these you need.

  • Custom systems

    Process sound, tooling isn't

    Software shaped around how the business actually works.

  • Agents and automation

    Repetitive work, done by hand

    Wired into the tools you already run on, built to be handed over.

  • Architecture and integration

    Systems that don't talk

    Deciding what to build, what to buy, and what to leave alone.

Our work

Twelve systems across nine industries: energy, manufacturing, lending, plantation, transport, e-commerce, hospitality, property and advertising.

  • Mobility marketplace

    Ground transport across worldwide operators: a supply platform, an admin portal for operations, and a mobile app.Ground transport operator · 2026    One booking flowride-hail, chauffeur, rental
    • More than 100 bookings a month and growing, from a mid-2026 launch.
    • More than 30 suppliers connected, in markets worldwide.
    • Ride-hailing, chauffeur and car rental handled in one booking flow instead of three supplier processes.
    • Every environment reproducible from Terraform across AWS, Cloudflare and Google Cloud.
    Role
    Team of eight
    Stack
    SvelteKit, Capacitor, Hono, PostgreSQL
    Read the full study →

  • The loan book

    Salary-deduction lending across three parties: the agent who originates, the lender who funds, and the co-operative that collects.Salary-deduction lender · 2026    Reconciliation built inAngkasa and the bank
    • Reconciliation against Angkasa and the bank cut from two weeks of work by two people to a routine inside the system.
    • More than 500 loans under management, live since mid-2026.
    • Agents, lender and co-operative working in one system instead of exchanging spreadsheets.
    Role
    Team of three
    Stack
    Next.js, FastAPI, PostgreSQL, Celery, Redis
    Read the full study →

  • The field foreman

    Durian plantation management: soil and weather readings, task assignment, and labour cost per zone.Wastech · 2020    Works with no signaloffline-first in the field
    • Reporting to the owner cut from weeks or months to same-day.
    • Deployed across two plantations, each divided into 10 to 20 zones.
    • More than 50 connected irrigation devices reporting soil moisture, weather and machinery hours.
    • 20 to 30 field workers running their day from the app, fully offline.
    • Irrigation scheduled against actual rainfall rather than a fixed routine, cutting water and the diesel used to pump it.
    Role
    Led, team of five
    Stack
    Flutter, C#, ASP.NET MVC, MySQL
    Read the full study →

  • Autonomous front desk

    The email channel of a ground transport marketplace, run end to end on Claude at a fixed monthly cost.Ground transport marketplace · 2026    Agents work the inboxoutreach, support, bookings
    • Averaging one new supplier onboarded a day, from first contact to trading.
    • Responses inside 30 to 60 minutes.
    • Outreach, customer support and inbound booking requests all handled on one channel by the same agents.
    • The same workflow at today's volume would take around 15 people to run by hand. It runs with three.
    • Cost is a fixed monthly subscription rather than a metered API bill that rises with every message.
    Role
    One engineer
    Stack
    Claude, React, Node.js, SQLite
    Read the full study →

  • Manufacturing ERP

    Custom ERP for a steel manufacturer that had outgrown AutoCount and could not customise it any further.Steel manufacturer · 2024    Production to accountsno reconciliation between
    • Replaced AutoCount, which could not be customised further to fit the production process.
    • Production, procurement, inventory and accounting run in one system rather than as separate exercises reconciled by hand.
    • The production process is modelled as the client actually runs it, rather than as the package required it to be described.
    Role
    Team of seven
    Stack
    C#, .NET, MySQL
    Read the full study →

  • Game top-up store

    In-game currency and digital goods sold at volume, with a storefront, a wallet and a transaction engine behind it.Dynasty Game Store · 2024    Promotions without a developerchanged in the CMS
    • Around RM10 million in transactions a month, without degrading under load.
    • Products, banners and promotions are changed by administrators through the CMS rather than through engineering.
    • Game providers are added through a modular API layer, so a new title is an integration rather than a rebuild.
    • Partner access runs through a documented OpenAPI layer with per-account keys and IP allow-listing.
    Role
    Team of four
    Stack
    C#, ASP.NET MVC, SQL Server
    Read the full study →

  • Grid forecasting

    Planning platform for an electricity distribution network, covering substations, feeders and switches across the grid.Electricity distribution · 2023    Demand 20 years outagainst assets in the ground
    • Planning moved off shared Excel files onto one platform, serving around 100 internal users: planners, management and technical staff.
    • Covers hundreds of substations and more than a thousand feeders, switches and other assets.
    • Runs many more forecast simulations than the previous process could, and produces more accurate ones. The forecast cycle is still measured in months.
    • Twenty-year equipment-change predictions shown on a map, with approvals routed through the hierarchy by email notification.
    Role
    Team of six, in a 15-person programme
    Stack
    Angular, C#, .NET, PostgreSQL
    Read the full study →

  • In-car advertising

    Android kiosk app for in-car advertising, locked to a single app so a driver cannot exit it.In-car advertising operator · 2022    Screens run centrallyunattended, in-vehicle
    • Deployed to 100 tablets, in a fleet run in partnership with an airline providing airport transfers.
    • Screens stay locked to the app in the field; drivers pause and resume behind a PIN instead of exiting it.
    • What the fleet is showing is changed centrally rather than vehicle by vehicle.
    Role
    Team of five
    Stack
    Angular, Kotlin, C#, .NET, SQL Server
    Read the full study →

  • Live-selling commerce

    nopCommerce customised to how each retailer already sells, across more than ten deployments in six sectors.Retail and marketplace operators    Live comments become orderswith a payment link back
    • More than ten clients over five years, across six retail sectors.
    • Live-selling order capture built once and then deployed to five separate clients.
    • Comments on a Facebook live stream become real orders with payment links, instead of someone transcribing them after the stream ends.
    Role
    Teams of two to four
    Stack
    nopCommerce, C#, ASP.NET, SQL Server
    Read the full study →

  • Back of house

    Procurement and inventory for a hospitality group running more than 100 outlets and around ten central kitchens.Hospitality / F&B operator · 2020    Costing, not estimatingstock, transfers and waste
    • Replaced WhatsApp and spreadsheets as the system of record for procurement.
    • Runs across more than 100 outlets and around 10 central kitchens.
    • Several thousand SKUs, with stock takes, inter-outlet transfers and food waste all recorded in one place.
    • Monthly usage, cost and quantity reporting, so costing is calculated rather than estimated.
    Role
    Team of seven
    Stack
    Flutter, C#, ASP.NET MVC, Python, SQL Server
    Read the full study →

  • The estate office

    Visitor management and property administration, run as one platform across a property operator's customer sites.Property management operator · 2022    Guardhouse to billingone resident record
    • Resident and customer registration is materially faster than the manual process it replaced.
    • Billing runs automatically each cycle instead of being prepared by hand.
    • The same resident record serves the guardhouse and the invoice, rather than being kept twice.
    Role
    Team of four
    Stack
    Flutter, C#, .NET, SQL Server
    Read the full study →

  • Wallet loyalty

    Loyalty platform for merchants: programme management, subscription billing, a developer API, and POS, email and CRM integrations.Loyalty platform operator · 2026    The card updates itselfin Apple and Google Wallet
    Role
    Team of three
    Stack
    Node.js, Express, Prisma, PostgreSQL, Google Wallet, Apple Wallet
    Read the full study →

How we work

The engineering happens inside your operation rather than at arm's length from it. In practice that is four things.

  • We work from the operation, not from a brief

    Mostly remote, in your tools and channels and alongside the people doing the job, and on-site when the process needs watching in person. Either way what we build from is the work itself, not a requirements document someone else wrote.

  • The engineer who maps it leads the build

    Whoever watched your process directs what gets built from it, so nothing is handed to a team you never meet and rebuilt from a written spec. Staying small and senior is what makes that possible.

  • You see it in increments

    Working software you can look at while it is being built, so a change of direction costs a week rather than a quarter.

  • You can take the source code

    Ask for it and the source is yours, so whoever runs the process afterwards is not tied to us. We would rather be brought back than depended on.

Engagement models

  • Drag to turn and tilt

    builds once, then holds

    Fixed price

    Well-defined projects with clear scope

    Predictable cost and timeline, delivered against milestones. Suits an MVP, a website, or a specific feature build.

  • Drag to turn and tilt

    keeps adding, always fits

    Time & materials

    Evolving projects with changing requirements

    You pay for hours worked and can see where they went. Suits long-term product development where the scope will move.

  • Drag to turn and tilt

    used up, then renewed

    Monthly retainer

    Ongoing support or a dedicated team

    A fixed block of hours each month at a lower rate, with priority response. Suits post-launch support and steady improvement.

Not sure which fits? Take the free 30-minute consultation and we will tell you which one your project actually needs.

Contact

Tell us what is slowing the business down.

If it is a fit we will say so, and if it is not we will tell you that too.

Goes straight to cheefi.ng@overcode.io, and you get a copy.