What does Epochal Tech build?

Epochal Tech is a Malaysian software studio that builds AI automations, SaaS MVPs and production backends, alongside custom websites. Current work includes a live retail pre-order platform, a multi-tenant pre-order SaaS and a daily AI video pipeline.

Can AI automations run without large API bills?

Often, yes. Where quality allows, steps run on local open models through Ollama, and paid APIs such as OpenAI or Claude are used only where they clearly do better. The daily video pipeline runs mostly on local models for that reason.

Why is the AI inside the workflow instead of running it?

A fixed workflow with an LLM inside a step is predictable, testable and cheap to debug. Letting a model decide the whole flow adds cost and failure modes that most business processes do not need.

Who owns the code and the accounts?

You do. Code lives in your repository and services run on your accounts, with documentation at handover, so there is no lock-in.

How is a project priced?

Per project, in fixed milestones agreed after a short discovery call. Each milestone ends in a working demo before the next one starts.

Automation & MVPs

AI automations and SaaS MVPs
that run in production.

Epochal Tech builds AI-powered workflow automations, SaaS MVPs and the backends behind them for startups and small businesses — not demos that break the week after handover. Every system below is running today, in production or as a public demo.

Discuss your project See the work

What I build

Automation that removes the busywork, products that ship.

AI & workflow automation

n8n workflows with an LLM inside a step, not in charge of the flow: document and email processing, lead capture, reporting and notifications to Telegram, WhatsApp or email.

SaaS MVPs

From idea to a product people can pay for: authentication, multi-tenant data, Stripe subscriptions, an admin console and a public demo.

Production backends

FastAPI and PostgreSQL services built to survive real traffic: row-level security, queues, idempotent jobs, replicas, connection pooling and tested backups.

Case studies

Real systems, all still running.

Retail pre-order & operations platform

Solo architect & full-stack developer · Live in production

Problem: A Malaysian retail and café chain ran pre-orders, café reservations, order confirmations and counter sales on four separate systems plus manual WhatsApp follow-ups, which meant retyped orders and double bookings on limited drops.
What I built: All four merged into one platform behind a single gateway, with no downtime for staff. Redis slot holds and queues stop overselling on hot drops; every database has a streaming replica, connection pooling and offsite backups with tested restores.

FastAPI
React
PostgreSQL
Redis
PgBouncer
Docker
Cloudflare

Préordre: multi-tenant pre-order SaaS

Founder & sole engineer · MVP with a public demo store

Problem: Small retailers take pre-orders through DMs, Google Forms and spreadsheets. They lose orders and cannot see demand before they buy stock.
What I built: A single-client system turned into a self-serve SaaS. Each merchant gets a branded storefront and an operator console for products, catalogue, collection and walk-in orders. Stripe handles subscriptions, and PostgreSQL row-level security keeps every merchant’s data isolated even if the application has a bug.

FastAPI
React
PostgreSQL RLS
Stripe
Docker

Open the live demo store

preorder.epochaltechmy.com/demo

Daily AI video pipeline

Designer & builder · Running daily

Problem: Publishing a video every day means hours of scripting, visuals, narration, editing and uploading, and paid AI APIs make that expensive at daily volume.
What I built: A scheduled pipeline that does all of it: a local LLM writes the script, local image generation draws the scenes, local text-to-speech narrates and FFmpeg renders. n8n starts it every morning, uploads to YouTube and reports to Telegram. Re-runs never double-post, and local models keep most runs near zero cost.

n8n
Ollama
ComfyUI
Piper TTS
FFmpeg
YouTube API

CRM lead follow-up automation

Designer & builder · Public demo

Problem: Website enquiries sit in an inbox until someone copies them into the CRM, decides how urgent they are and remembers to follow up. Slow replies lose the lead, and open deals go quiet without anyone noticing.
What I built: An n8n workflow that handles an enquiry in seconds: it saves the contact in HubSpot without creating duplicates, scores the lead hot, warm or cold with a local LLM that reports how sure it is, opens a deal at the matching stage, sets a follow-up task and drafts a reply for a person to approve. A scheduled check finds deals that have gone quiet and creates reminders. The public demo uses a fictional business and sample customers, and never sends anything.

n8n
HubSpot CRM
FastAPI
Ollama
SQLite
Docker

Try the live CRM demo

Self-hosted server monitoring dashboard

Designer & builder · In daily use (internal)

Problem: One server runs every production system as Docker containers. Finding out that a container had stopped, that memory was running out or that a site was being probed meant logging in and checking by hand.
What I built: A private dashboard for the whole server: CPU, memory and every container updated live over WebSocket, container logs on demand, and an alert when a container stops that clears itself when the container is back. It also reads the request log of a production gateway to flag server errors, blocked staff-page probes and traffic spikes, and charts the daily YouTube channel figures that an n8n workflow collects. It is reachable only over a private network, never from the public internet.

FastAPI
PostgreSQL
Next.js
WebSocket
Docker
n8n

How I work

A founder’s view, an engineer’s hands.

Discovery first

A short call to agree what success looks like before any code is written.

Fixed milestones

Each milestone ends in a working demo you can click, not a status report.

You own the code

Your repository, your accounts, documented handover. No lock-in.

Secure by default

Secrets stay out of code, inputs are validated server-side, access is checked on every request.

Remote, GMT+8

Based in Malaysia, with working hours that overlap Europe and the US east coast.

Stack

Python
FastAPI
TypeScript
React
Node.js
PostgreSQL
Redis
Docker
n8n
OpenAI / Claude APIs
Ollama
Stripe
Supabase
Cloudflare

Questions

Straight answers.

What does Epochal Tech build?: Epochal Tech is a Malaysian software studio that builds AI automations, SaaS MVPs and production backends, alongside custom websites. Current work includes a live retail pre-order platform, a multi-tenant pre-order SaaS and a daily AI video pipeline.
Do you work with clients outside Malaysia?: Yes. Work is fully remote from Malaysia (GMT+8), with working hours that overlap Europe and the US east coast, and communication in English or Chinese.
Can AI automations run without large API bills?: Often, yes. Where quality allows, steps run on local open models through Ollama, and paid APIs such as OpenAI or Claude are used only where they clearly do better. The daily video pipeline runs mostly on local models for that reason.
Why is the AI inside the workflow instead of running it?: A fixed workflow with an LLM inside a step is predictable, testable and cheap to debug. Letting a model decide the whole flow adds cost and failure modes that most business processes do not need.
Who owns the code and the accounts?: You do. Code lives in your repository and services run on your accounts, with documentation at handover, so there is no lock-in.
How is a project priced?: Per project, in fixed milestones agreed after a short discovery call. Each milestone ends in a working demo before the next one starts.

Start

Have a process worth automating?

Describe the workflow or product you have in mind. You will get a reply within 24 hours on working days with questions or a first plan.

Discuss your project Chat on WhatsApp

AI automations and SaaS MVPsthat run in production.