AI & Machine Learning

AI that does real work inside your ERP.

We build practical AI for business: document extraction that posts vendor bills into Odoo, demand forecasts your planners can question, quality inspection on the line and answers grounded in your own records. Scoped like engineering, priced per phase, with a human review step wherever money moves.

Find my first AI use caseSee what we build
  • Document extraction into Odoo
  • Demand forecasting
  • Vision-based quality inspection
  • Runs on your own servers
  • Human review where it matters
Models, code and training data owned by you
Scope and price per phase
Delivery languages
Reply time on enquiries

What we build

AI applied to work your team does every day

We do not sell AI by the buzzword. Each of these starts from a task someone does by hand today, and each is measured against how that task was done before.

Document extraction

Vendor bills, receipts and delivery notes read into structured Odoo records, matched to purchase orders, with low-confidence fields routed to a review queue.

Demand forecasting

Forecasts per SKU and branch built from your own sales and stock history, feeding Odoo reordering rules and MRP instead of another spreadsheet.

Quality inspection with vision

Camera-based defect detection and counting on the line, writing results into Odoo Quality checks so a bad batch is stopped before it ships.

Anomaly detection

Unusual expenses, duplicate payments and odd stock movements flagged for a person to review, learned from your normal patterns.

Answers from your own data

Ask a question in plain Thai or English and get an answer grounded in your documents, manuals and Odoo records, with the sources cited.

Scoring and segmentation

Lead scoring for CRM follow-up and RFM segmentation for campaigns, so sales calls the right ten people first.

Workflow automation

Multi-step automations for routing, approvals and follow-ups that check their own results, log every action and stop for approval where it matters.

Private deployment

Open-source models running on your own cloud or hardware where the data is sensitive, so nothing leaves servers you control.

AI inside Odoo

Start where the payback can be measured

The fastest AI wins sit inside processes you already run. A vendor bill that posts itself saves minutes every single day, and a forecast that trims overstock shows up in the balance sheet. We wire AI into Odoo records and workflows in separate modules, and our own TechAfterMe Dashboard app already ships optional one-click AI insights that use an API key you control.

  • Extraction results land as draft Odoo records, never posted without a review rule you set
  • Forecasts feed reordering rules and MRP, where they change real purchasing decisions
  • Every automated action is logged, with an approval step wherever money moves
  • Sensitive workloads can run entirely on your own servers with open-source models

  1. Measure

    We time and cost the manual process first, so the result has a baseline.

  2. Prototype

    A working proof on your real documents and data, at a fixed price.

  3. Integrate

    Wired into Odoo as a separate module, with tests and guardrails.

  4. Review

    Accuracy is tracked in production, and exceptions go to a human queue.

Where we start

Three teams AI helps first

A finance team buried in vendor bills

Hundreds of PDFs keyed in by hand each month become draft bills matched to purchase orders, with people reviewing exceptions instead of typing everything.

A planner forecasting in Excel

Purchase quantities guessed from last year become per-SKU forecasts feeding reordering rules, with the planner approving and correcting, not copying cells.

A factory inspecting by eye

Spot checks that miss defects on a tired afternoon become camera checks on every unit, logged against the work order in Odoo Quality.

Industries

The same methods, shaped to each sector

The models matter less than the workflow around them. These are the applications each sector usually pays back on first.

How we work

From a candidate list to AI in production

  1. 1. Discover

    We shortlist use cases with a measurable payback, and we say plainly when a report or a plain script is the cheaper answer than AI.

  2. 2. Prototype on your data

    A fixed-price proof on your real documents and history, measured against the manual baseline before you commit to a build.

  3. 3. Build for production

    The model becomes a monitored service with tests, an Odoo module with zero core overrides, and guardrails written into the scope.

  4. 4. Deploy with review

    Automation starts on a slice of the work with a human queue for exceptions, and widens only as the accuracy numbers earn it.

  5. 5. Improve on evidence

    Accuracy and drift are tracked in production, retraining is scheduled from real corrections, and each next phase is quoted separately.

Our stack

Proven tools, the smallest model that works

  • Python
  • PyTorch
  • scikit-learn
  • OpenCV
  • Tesseract OCR
  • pgvector
  • PostgreSQL
  • FastAPI
  • Label Studio
  • Docker
  • Grafana
  • Odoo

Our commitments

AI promises without the hype

We will not quote an accuracy percentage before measuring on your data, and we will not promise headcount savings we cannot see. These are the commitments we sign.

Find my first AI use case
Review step wherever money moves
Price per phase, agreed in writing
Written status note
Reply time on business days

FAQ

Questions clients ask first

Is our business ready for AI?

You need a real problem and some history or sample documents, not a data-science team. A discovery call tells us whether the payback is there, and sometimes the honest answer is that a better report or a plain script solves it cheaper. We say so when it is.

How do you keep our data private?

Access is scoped to the task, data is encrypted in transit and at rest, and sensitive workloads can run open-source models entirely on your own servers. Your data is never used to train anything shared with other clients, and in Thailand we design retention and access around PDPA duties.

Which AI models do you use?

The smallest one that solves the problem. Classical machine learning covers most forecasting and scoring, vision models handle inspection, and text tasks use hosted or self-hosted open-source models chosen per case for accuracy, cost and privacy. The trade-off is written down before we build, and you can veto any external service.

How accurate will it be?

We do not know until we measure on your data, and neither does anyone else. The prototype phase reports accuracy on your real documents or history, the review queue catches what the model is unsure about, and automation only widens as the numbers prove out in production.

Can AI work inside our existing Odoo?

Yes. We integrate with Odoo 16 through 19 as separate modules with zero core overrides, so upgrades stay possible. Extraction creates draft records, forecasts feed reordering rules, and our TechAfterMe Dashboard app adds optional one-click AI insights using an API key you control.

What does an AI project cost?

Discovery is a fixed fee, the prototype is a fixed price against a written goal, and production phases are quoted separately like any engineering work. Cloud and any external model costs run on your own accounts with no markup, and there is no per-seat license to us.

Start with one use case

Tell us the task your team wishes would do itself

Describe the manual work, roughly how often it happens and where the data lives. You get a reply within one business day, and after a discovery call, a written plan with a fixed price for a prototype on your own data.

  • Reply within one business day
  • Fixed-price prototype before any big build
  • Your data, models and code stay yours

Prefer to talk first? Call us or WhatsApp us.

Your details stay private. We reply within one business day.