Sydney-based AI transformation for fresh produce, food distribution, warehousing and logistics

We put engineers on your floor and AI into your daily operations.

Transcript
  1. Orders arrive as emails, PDFs, photos, faxes.
  2. AI reads each one, whatever the layout.
  3. Customer, date and every line fill in.
  4. Red means fix it. Amber means take a look.
  5. Your team checks, then sends with one click.
  6. Then the next, and the next.

Our engineers work on site with your team and deliver digital transformation, AI efficiency and a data warehouse your people use every day.

In production TOPO OrderSnap has run in production at the distributor we build for since early 2025.

Our products

Five products, ready-made parts our engineers fit into your operation: an AI agent in your team's WhatsApp group, AI order entry, field sales, an enterprise data hub and the next-generation platform.

All products

Highlights

A few things we have built, a few seconds each. For how they work, ask us for a walkthrough.

  • Goals that move with the business

    Set the year's sales goal and watch it flow into every month and every rep; through the year, everyone sees who is ahead and who is behind.

    Ask for a walkthrough

  • The warehouse floor in live 3D

    Racks coloured by how full they are and vehicles on the move, fed read-only by the systems already on the floor.

    Ask for a walkthrough

  • See what a failure touches

    When one table fails, every report, app and team downstream lights up, layer by layer in 3D.

    Ask for a walkthrough

  • AI agents on watch

    A group of AI agents watches the data, flags what changed and proposes the fix when something fails.

    Ask for a walkthrough

Where the hours go

  • Orders keyed in by hand

    Emails, PDFs, photos and faxes arrive all day. Someone re-types every line, and the mistakes show up at the dock.

  • Systems that don't talk

    Sales, stock and invoicing sit in separate systems. One clear number means a spreadsheet and a late night.

  • Too risky to switch

    The old system is slow, but it ships orders every day. A bad cut-over could stop the business.

  • AI, but where to start?

    Everyone talks about AI. Few say which job to start with, or who will make it work on your floor.

One solution, from customer order to management report

Seven links in one chain. Each shows which service line it belongs to.

  1. Customer orders

    Email, PDF, photo, fax or spreadsheet. Customers keep ordering the way they do now.

    In productionAI efficiency
  2. Order intake AI (TOPO OrderSnap)

    AI reads each order and completes it. Your team reviews it and sends it with one click.

    In productionAI efficiency
  3. Your ERP / WMS

    Orders post into the system you run today, so nothing has to be replaced to start.

    In productionDigital transformation
  4. Warehouse floor

    Boards refreshed through the shift, crew rostering and data pulled from the systems you already have, without new equipment.

    DeliveredDigital transformation
  5. Data warehouse (TOPO DataHub)

    Your ERPs and other systems in one cloud data warehouse: on-premise ERPs fully refreshed by about 7 am, open orders and sales updated through the day.

    In productionData warehouse
  6. AI applications

    TOPO CRM on iPhone and web, and TOPO Ask, the AI agent in your team's WhatsApp groups: the daily business briefing and plain-language questions about your numbers. Next-day purchasing forecast: v1 delivered July 2026.

    In productionAI efficiency
  7. Management reporting

    One set of numbers for the whole business. Each measure is defined in the warehouse, not in each report.

    In productionData warehouse

Three service lines, one team

Digital transformation

Replace paper, spreadsheets and re-keying with systems your people use every day, without ripping out what already works. Our engineers have delivered on site in large automated and manual distribution centres.

  • See the floor without new equipment: read-only feeds from your existing automation into boards refreshed through the shift, showing jams, busy and critical zones, and picking.
  • Rostering for the crews on a large distribution-centre floor: planned the night before, a supervisor roll call at the start, updated through the shift, everyone clocked off with hours recorded, every change logged.
  • Data from systems with no API, pulled on a schedule with your own read-only accounts: unattended and auditable.
  • TOPO NEXT: replace your ERP one module at a time, each proven side by side with what you run today before it switches over, starting with pricing.
  • Odoo 17 and 18 implementation. We run Odoo ourselves.

AI efficiency

Take the typing and chasing out of the day with LLM applications, AI agents and agentic automation. AI does the repetitive work; your people check it and decide.

  • TOPO OrderSnap for order entry: AI reads customer orders in any format and enters them as complete orders, and your team reviews and sends each one with one click.
  • TOPO CRM for field sales, on iPhone and web: this week against target, and customers who should have ordered and haven't.
  • TOPO Ask, an AI agent in your team's WhatsApp group: ask about sales in plain words, and it writes the query, runs it on your data warehouse and answers in the chat. It also posts the daily briefing, counting orders not yet invoiced.
  • Next-day purchasing forecast, v1 delivered July 2026, for the categories where it beats 'buy what you bought last week'.
  • TOPO NEXT: our next-generation platform, from the customer's order to delivery, ready to demonstrate and tailored to your operation.

Data warehouse

All your numbers in one place, each measure defined in the warehouse rather than in each report, updated from early morning and through the day.

  • On-premise ERP databases and a cloud system loaded into a cloud data warehouse: the on-premise ERPs fully refreshed by about 7 am, with the open-order book every hour; the cloud source updated every hour; same-day sales several times a day.
  • Three layers, from raw to semantic to one shared view. Each measure is defined in the warehouse, not in each report.
  • 778 days of history (1 Jul 2024 – 17 Aug 2026) backfilled when we connected a new source system to the shared warehouse, with 0 failed loads.
  • Views we write are kept in version control; views maintained in the console are backed up there. A health-check script covers every source.
  • TOPO DataHub: ERP and database sync, file sync and web data collection, gathered read-only into one data lake with monitoring and KPI dashboards on top.

Results

We provide IT services and product development to a fresh-produce distributor. TOPO OrderSnap, TOPO CRM, TOPO Ask with the daily briefing, and the data warehouse run in their production every day. Every number comes with its window, and estimates are marked.

Estimate
About 9 seconds

for TOPO OrderSnap's AI to read an order and enter it, next to about 50 seconds to read and key a 7-line order by hand (our estimate)

TOPO OrderSnap: the AI step only, from reading the order to the completed entry, estimated for the current version; your team's review comes after it. By hand: estimated from published keystroke and reading times, not timed; 25–90 seconds depending on the person and the order.

In production
Measured
99.74%

of invoice lines priced to the cent by TOPO NEXT's new pricing engine in a 30-day replay against the live system (99.40% by value)

Every sales line at one warehouse, delivered 16 Sep – 15 Oct 2026 (returns excluded).

Measured
31% lower

forecast error than 'buy what you bought last week', for our next-day purchasing forecast

MAPE 37.2 vs 54.1; 15 market-day backtest 27 Jun – 14 Jul 2026, value-weighted, daily-buy categories, one warehouse. v1 delivered July 2026.

Delivered
Measured
778 days

of history backfilled when we connected a new source system to the shared warehouse, with 0 failed loads

1 Jul 2024 – 17 Aug 2026, one source system.

In production

How new technology lands

New tools earn their place on real work, with a person in charge.

  • Multimodal AI reads the orders

    TOPO OrderSnap uses AI to read orders in any layout, typed, scanned, photographed or faxed, and enter them.

    In production
  • Models compared on real work

    We compare candidate models on real work before choosing one. For example: TOPO OrderSnap's default model (April 2026, four models on real customer orders) and the model behind the CRM's meeting preparation and summaries (July 2026, four models on the real meeting-preparation prompt).

    In production
  • AI coding agents, supervised by engineers

    Engineers set the plan and accept the result; AI coding agents do the work in between and let each engineer deliver more, which is what makes a forward-deployed engineer affordable for a mid-sized business. Example: TOPO NEXT's new pricing engine, 41 reviewed commits and 334 passing tests between the afternoon of 16 Oct and early 17 Oct 2026 (the first ten steps of the first plan). On this project, every change passed independent AI review and the full test suite before it landed.

  • A fix replayed on real data before release

    Before the October 2026 TOPO OrderSnap customer-recognition fix went live, it was replayed against two months of real lookups.

    In production
  • Local AI on your own GPU server

    For documents that must stay in your building, the AI runs on a GPU server on your premises.

How we work: reuse first, then build what is yours

Our team has warehousing and logistics experience, and specialists in both. So we start from what already works and build only what is yours alone.

  • First choice

    Reuse what already works

    Where your operation works like the ones we know, we reuse our proven products and know-how directly: TOPO OrderSnap, TOPO CRM, TOPO Ask and TOPO DataHub, set up for your systems. It is the fastest and lowest-risk way in.

  • Shadow your people, then deploy AI agents

    Where your work is different, we shadow your staff on site, document how they do the job and extract the skills behind it, then deploy those skills as AI agents that work alongside your team, with a person checking the result.

  • Engineers on your floor

    What only your business does, forward-deployed engineers build on site, with your people and your data.

However the work is split, every project runs in the same four steps, and you can stop after any of them.

  1. Learn the floor

    We start where the work happens: the order desk, the cold room, dispatch. We watch, ask, and look at your systems read-only.

    You can stop here and keep the written findings.

  2. Build the first working piece

    We pick the job that costs the most hours and build the first piece that works on your real orders and records.

    You can stop here and keep what was built.

  3. Deploy where you need it

    It runs side by side with the current way first, in shadow, and each release comes with a rollback point.

    You can stop here. Nothing switches over until the numbers match.

  4. Run and improve

    We keep it running, watch the daily checks and keep improving it with your team.

    You can stop at any time. The code and data we built for you can be handed over in full.

Our data promise

  • Your servers or your cloud

    By default we deploy on your servers or in your own cloud account; we can also host it for you.

  • Not used to train AI models

    For your deployment we use paid AI services whose terms exclude training on your data. For documents that must stay in your building, the AI runs on a GPU server on your premises.

  • Separate, under NDA

    Each client gets a separate database or project, under a confidentiality agreement.

  • Full handover

    When the project ends, the code and data we built for you can be handed over in full.

We have worked with

Talk to us

Tell us which job takes the most hours. Email or call. We start with a conversation and, if it makes sense, a visit to your floor.