P-Techx · Production-ready AI for manufacturing, energy, and the enterprise

AI that actually
connects to
your reality.

We integrate AI into real systems — production lines, backend databases, daily workflows. Not isolated demos.

Book a 30-minute consultation See the proof
Scroll
Ten years.

Of team experience building the ERP, SCADA, and CMMS systems your AI actually has to talk to.

Six sectors.

Where the team's production systems run: automotive, energy, retail, healthcare, finance, and enterprise operations.

Still live.

Our longest continuous AI engagement is two years and eight months old, and running today.

The enterprise AI reality check

Most enterprise AI never leaves the demo.

Typical AI vendors
The P-Techx reality
Typical AI vendors

Endless proof-of-concepts that die in production.

The P-Techx reality

Systems engineered to survive production and scale.

Typical AI vendors

Rented, generic APIs that do not understand your business.

The P-Techx reality

Custom models trained specifically on your proprietary data.

Typical AI vendors

Vague promises of increased efficiency.

The P-Techx reality

Strict, trackable metrics tied directly to ROI.

Typical AI vendors

Isolated dashboards disconnected from core operations.

The P-Techx reality

Deep hooks into your existing ERP, SCADA, and PLCs.

Proof 01 · Tier-1 automotive supplier · vision & OCR

93.7%

Field-level accuracy across 689 factory images and 7,798 fields, at 0.8 seconds each.

Two models we trained ourselves: one sorts the tag among 24 supplier types at 99.2%, the second reads the fields — 94.7% on the critical ones. 18 days of scored iteration took it up from 76.8%.

Talk through this engagement
A supplier production label mixing Japanese, English, Thai, and handwritten fields.
SUPPLIER TAG · 4 SCRIPTS · PRINTED + HANDWRITTEN

Proof 02 · Same client, second system · production traceability

60,000

What one production line turns out in a day — every piece of it now traceable, for the client who came back after the OCR system.

An edge service reads the line's PLCs and inspection cameras every second, counts good and rejected parts per lot, and prints the traceability label itself the moment a lot completes. Scan any QR and the system walks the part's history backwards through every station — washing, plating, assembly, packing. In production, with its own PLC emulator so upgrades are tested without touching the line.

Talk through this engagement

Proof 03 · Energy & solar management

32mos

Two years and eight months of unbroken anomaly detection across a fleet approaching a thousand sites.

Every item on the urgent list carries a price — one underperforming inverter, about ฿21,600 a month, flagged before anyone notices the dip. It doesn't report abnormal; it reports what abnormal costs.

Talk through this engagement
Solar fleet dashboard showing site totals, an urgent-inspection list priced per month, and a 30-day generation curve.
FLEET DASHBOARD · CLIENT NAME AND SITES REDACTED

Proof 04 · LINE AI assistant · membership platform

22actions

Back-office actions the assistant can take for the member it's talking to — every answer drawn from that member's own records, live on LINE.

It knows who it's talking to. Bound to each member's LINE identity, a question about "my orders" or "my team" is answered from that member's data and nobody else's — not a script, and never someone else's records. High-risk actions still stop for a human, and the whole thing runs at about ฿121 a day, tracked live on the cost console.

Talk through this engagement
The assistant's reply card in LINE showing a member's team structure, names cropped.
LIVE IN LINE · MEMBER NAMES CROPPED
Admin console showing daily AI cost split by model, cache hit rate, and token mix over two weeks.
COST CONSOLE · SPEND PER DAY · CACHE HIT RATE

Proof 05 · Thai product search · capability benchmark

94%

Recall@10, up from 50% — semantic queries at 89%, typo tolerance at 85%, both from zero.

Keyword search fails silently in Thai. We built the 56-query eval harness first, then a hybrid retriever — Postgres full-text and pgvector, fused with RRF.

Capability benchmark on our own eval harness — 56 queries, reproducible on request.

Talk through this engagement

What one bad inverter costs

฿21,600/mo

What a single underperforming inverter quietly drains — found and priced before anyone notices the dip. Now multiply by a fleet approaching a thousand sites.

Capabilities · 01 — 04

Four capabilities.
One integration layer.

01 — Core practice

Enterprise knowledge. AI assistants.

Permission-aware AI answering from your own company data, with an exact citation behind every answer.

02

Visual inspection.

Custom models reading complex manufacturing tags, and QA running where the parts are.

03

Autonomous workflows.

Agents doing real work across platforms, held by human approval gates.

04

Intelligent analytics and AI search.

Hybrid semantic search that reads native-language intent — so customers find the thing.

Surviving the PoC

Four stages.
Three gates.

01

Discovery 1 — 2 weeks

We define the evaluation set from your real data before anyone writes a model.

Gate 1 · Measurable goals agreed →
02

Proof of Concept 2 — 4 weeks

A working prototype and the first evaluation report, scored against that baseline.

Gate 2 · Hard metrics met →
03

Pilot 4 — 8 weeks

Real users, cost guardrails on, monitoring live.

Gate 3 · Real-world ROI proven →
04

Production Ongoing

Continuous deployment, alerting, an SLA, and retraining loops as your data moves.

SLA · Alerts · Retraining loops

Miss the metrics at any gate and we stop. Kill criteria are set on day one.

Why it works · 01 — 04

Why our AI succeeds
where others fail.

01 — Foundation

Engineering first. AI second.

We build the ERP, SCADA, and CMMS systems ourselves. That is why our AI integrates into them instead of sitting beside them.

ERP · SCADA · CMMS · PLC
02 — Measurement

The evaluation harness.

Your real data, scored against a fixed baseline for Recall@K, field accuracy, and latency. Hard numbers, not opinions.

BASELINE → SCORE → REPORT
03 — Engineering

Custom models, built to survive.

Trained on your data, not bent out of an off-the-shelf API. Dead-letter queues, retries, and alerting, because operations are chaotic.

TRAIN · DEPLOY · DLQ · RETRY
04 — Control

Your data never leaves your boundary.

SaaS, VPC, on-premise, or air-gapped — permissions live in the architecture, not in a prompt. Every call logged, credentials encrypted, high-risk actions gated on a human.

AIR-GAPPED · AUDIT TRAIL · HUMAN APPROVAL
Also delivered · 01 — 04

The systems behind
the six sectors.

Finance · in use

Accounting your non-accountants can run.

A multi-tenant accounting and ERP system for Thai businesses — Thai chart of accounts, tax and withholding built in, and a mobile app where AI reads the bill. Staff key nothing they don't understand; the accounting firm gets clean exports.

THAI CoA · WHT · AI-READ BILLS
Healthcare · production, now expanding

A public-health clinic, end to end.

Appointments, OPD, labs, medication, and referrals for a public-health clinic — ran in production, and now being rebuilt around tuberculosis care to Ministry of Public Health standards. Patient data never appears in our materials; demos run on mock records only.

OPD · LABS · MEDICATION · TB PROGRAMME
Industrial · delivered

From firmware up.

An STM32 Modbus gateway managing up to 128 pH sensors — firmware, closed-loop control, and the Windows app that configures it. When we say we work down to the metal, this is the metal.

STM32 · MODBUS RTU · 128 DEVICES
Retail enterprise · in production

The CMS behind a national retailer.

A deeply customised Strapi running content for a national retail platform — merchandisers compose product shelves, banners, and campaign pages from templates, preview in Thai and English, and publish on their own schedule. Media pipeline on S3 and CDN, deployed on AWS, built inside a twelve-person product team. 1,871 commits of it are ours to stand behind.

STRAPI · AWS · 1,871 COMMITS
The CMS page builder: two product shelves being arranged on a cart page, with a template palette and a draft-to-publish workflow.
PAGE BUILDER · DRAFT → PUBLISH · VIEW-ONLY ACCESS
Who we are

Four people who have
shipped this before.

P-Tech & Consulting Co., Ltd. was incorporated in 2025 in Bangkok. The company is new. The production experience behind it is not — the four of us spent the decade before it building the ERP, SCADA, and CMMS systems that our AI work now plugs into.

That is the whole team. No account layer, no offshore hand-off: the people who scope your engagement are the people who build it.

Registered entity
P-Tech & Consulting Co., Ltd.
Incorporated
2025
Based in
Bangkok, Thailand
Team
Four, all senior
  • Founder & solutions architect
  • AI / ML engineer
  • Platform engineer
  • Systems integration engineer
Get started

Let’s talk about
your reality.

Tell us your operational bottleneck. We’ll tell you where AI helps, how we’d measure it, and — crucially — where you shouldn’t use it at all. Free, 30 minutes.

Book the consultation
Contact

Send the brief.

A few sentences is enough. One business day for a reply, and no obligation to book a call.

Contact
Pratchaya Palee
Consultation
Free · 30 minutes

We use these details only to reply to this enquiry and to scope the work with you. We don’t sell them, share them outside P-Techx, or add you to a mailing list. Ask us at any time to see or delete what you sent.เราใช้ข้อมูลนี้เพื่อติดต่อกลับและประเมินขอบเขตงานเท่านั้น ไม่ขายต่อ ไม่ส่งออกนอกบริษัท และไม่นำไปเข้ารายชื่อส่งข่าว ขอดูหรือขอลบข้อมูลได้ทุกเมื่อที่ Saintentex@gmail.com