Inference Loops

AI consultancy in Phnom Penh

We help companies in Cambodia see where AI fits, shape the idea with you, and build what is worth building.

Talk to us Check your fit

What we can do for you

Most of our work starts with thinking, not code. We help you see clearly, decide well, and only then build.

AI opportunity review

We look at how work really gets done in your business and show you where AI pays off: what to automate first, what to leave alone, and what it will take.

Ideation workshops

Bring your team and a rough idea. Together we turn it into a clear concept, a first sketch and the questions to answer before anyone builds.

AI strategy and roadmap

Priorities, costs and risks for each idea, and a clear call: build it, buy it, or wait.

Data readiness check

What you record, where it lives, what is missing, and how to fix it before AI touches it.

Pilots and builds

A first working version, tested with the people who will use it. Then the full system, including Odoo ERP.

Team training

AI skills for your people in Khmer and English, built around their own work so the tools keep being used.

Bring us a question

These are the ones we hear most. Pick one to see how we would work on it.

Where would AI actually save us time?

How we would work on it

  1. Watch a normal week of work
  2. List the tasks that repeat
  3. Rank them by time saved and risk

Start with AI opportunity review

Ask us this

How we work with you

Every engagement is one turn of the loop. Short turns, real results, and each one makes the next one better.

  1. 1

    Discover

    We talk to your people and see how the work is really done today. No assumptions.

    What happens
    Interviews and time alongside the people doing the work. We look at your processes, tools and data as they are.
    You leave with
    A clear picture of where time and money go, and where AI could help.
  2. 2

    Ideate

    Workshops turn what we found into ideas worth testing, with the people who will use them.

    What happens
    Workshops with your team. We bring examples of what AI can do today and sketch ideas against your real work.
    You leave with
    A shortlist of ideas, each with its value, effort and risks.
  3. 3

    Decide

    We agree together what to build, buy or leave, and why. You get the reasoning, not just the answer.

    What happens
    We weigh each idea with you and check that your data is ready for it.
    You leave with
    A roadmap with priorities, costs and the reason behind each call.
  4. 4

    Build

    A first version quickly, tested on real work. Then the real system, if it earns it.

    What happens
    We build in short rounds, on Odoo ERP where it fits, and put each round in front of the people who will use it.
    You leave with
    A working pilot you can judge on results, and the full system if it earns it.
  5. 5

    Run and learn

    Results come back in and start the next turn of the loop. Each turn is shorter and sharper.

    What happens
    We watch how it is used, measure it against what we agreed, and train your team in Khmer and English.
    You leave with
    What worked, what to change, and the next question for the next turn.
Discover Ideate Decide Build Run and learn

Our approach: AI and real-world assets

AI runs the work. Clean records make it worth something.

Most of a business's value sits in real things: buildings, stock, kitchens, machines and people's time. AI can only run the work around them when the records are right. We build the systems that keep those records, so routine work runs itself and the value can be seen, checked and invested in.

  1. 1

    Structure

    One record for people, money and assets, instead of a spreadsheet in every office.

  2. 2

    Automate

    AI and workflows take over the routine: approvals, reminders, reports.

  3. 3

    Report

    Numbers anyone can check, every month, with a trail behind each one.

  4. 4

    Raise

    A business investors can read. Clean records are what they ask for first, for shares or for tokenised assets.

What you can count on

  • Built with AI from the first line

    AI works alongside us on every system, so we ship in weeks, not quarters.

  • Khmer and English

    Screens, documents and training work in both languages from day one.

  • Where people already are

    We meet teams in the apps they already use, like Telegram. No new installs, no training day.

  • Private by design

    Personal data is encrypted, every action is logged, and AI sees only what it needs.

  • Your data stays yours

    Systems can run on a private deployment that you control.

More from Inference

  • Products Coming

    Tools we build once and keep improving, by monthly access: a Khmer-first content engine and AI training for teams.

  • Community Next

    Open AI meet-ups in Phnom Penh, with CAFT members and fintech sponsors. A podcast later.

  • Research Ongoing

    Private and Khmer-language AI, and a security lab that tests our own systems first.

Inference Loops logo

Why "Inference"

In AI, inference is the moment a trained model meets something new and acts. For us it is the moment what people know becomes something that runs. We are based in Phnom Penh: Cambodia first, then the region.

Mission
Bring Cambodian businesses into the automation era: structured, automated and ready to grow.
Vision
Companies, people and assets in Cambodia running on AI they can trust. Built here first, then shared across ASEAN.

Writing

All writing
  1. The Privacy Leak Is Between the Agents: Why Output Audits Miss Most of ItSplitting one agent into a coordinator and workers is supposed to be the safer architecture. A 2026 benchmark shows it quietly moves the privacy risk somewhere your audits can't see. AgentLeak (arXiv 2602.11510) instruments seven communication pathways across 1,000 scenarios, five production models, and 4,979 execution traces — and finds that multi-agent setups do lower final-output leakage (27.2% vs 43.2% single-agent) but push total system exposure to 68.9%, because inter-agent messages leak at 68.8%. The blunt consequence: an output-only audit misses 41.7% of violations. Here's why the leak lives in the coordination channels, why you have to instrument them, and why that's an opportunity for teams close to regulated, local data.
  2. On-Device Agents and the Privacy Dividend: Why Local Isn't Yet PrivateRunning an agent on your own hardware is sold as a privacy win — the data never leaves the building. But 'local' and 'private' are not the same claim, and 2026 made the gap precise. AgenTEE (arXiv 2604.18231) places an agent's runtime, inference engine, and tools into independently attested confidential VMs on Arm's Confidential Compute Architecture, enforcing hardware isolation even on a device its own user might tamper with — at under 5.15% overhead. Amulet (arXiv 2512.07495) does the matching job for the model itself, shielding weights with TEE-backed obfuscation at 2.8–4.8x the latency of an unprotected model while running 8–9x faster than naive full-TEE inference. Here's why the real privacy dividend isn't 'it runs locally' — it's confidential, attestable execution — and why that mechanism favors Southeast Asia.
  3. We Named Our Company After This: Loop Engineering Is Eating Software DevelopmentThe man who built Claude Code says he doesn't prompt it anymore — he writes loops that prompt it. That single line named the paradigm shift the whole agentic era has been building toward: from one-shot prompting to persistent, feedback-driven loops. Here's why the prompt was always a placeholder, why the loop is becoming the unit of software work, and why a company called Inference Loops bet its name on exactly this.

Talk to us.

Three ways to start. Not sure which fits? Take the two-minute fit assessment.

hello@inferenceloops.com Phnom Penh, Cambodia