Consultant view
← Back to profile

Senior software engineer & founder · Available for advisory and build work

Tai Vu

I design and ship agentic infrastructure for solo founders and small teams. Same-day MVPs, four-instance n8n architectures, self-hosted AI content stacks. References: three of my own ventures run on systems I built.

Melbourne, AU  ·  Available for advisory & build engagements

tai@trendai.au · Book a 15-min discovery call

116 components in production
3 ventures running
15 years across roles
8 months solo to v1
01

What I build

01

Agentic infrastructure

Self-hosted operating systems for one-person companies — multi-LLM kernels, persistent memory on Postgres + pgvector, doctrine-as-code, dual-vault secrets, monitoring + backup as first-class concerns. Built TrendAI OS to run three businesses in parallel.

02

Production AI apps

Same-day MVPs on Next.js + Cloudflare Pages + D1, with Gemini OCR, Claude tool-use, and CF Access SSO. Idea-to-production measured in hours, not weeks. Reference: taikim-finance replaced a 3,997-row Excel log in one afternoon.

03

Workflow automation at scale

Four production n8n instances under one architectural doctrine — scheduler / automation / sandbox / lab, Postgres-backed, regression-monitored with 288 synthetic probes/day. Resumable-pipeline pattern with three-layer durability. Zero submissions lost in four months of production.

02

Selected work

TrendAI OS — agentic infrastructure

Self-built operating system for solo founders: 5+1 architecture layers — substrate (OCI + Cloudflare + Tailscale), multi-LLM reasoning kernel (Claude + Gemini + self-hosted Qwen), Postgres + Supabase memory, four n8n orchestration instances, Claude Code agent team, BUD operator cockpit. 116 components live in production. Solo-built over 8 months and documented as binding doctrines.

OCICloudflareAnthropicGeminin8nSupabaseTailscale

TrendAI Knowledge Engine — multi-source AI ingest

Four-workflow n8n pipeline ingesting YouTube transcripts, articles, and inbound email. AI-summarised with Gemini 2.5 Flash, persisted to Supabase, surfaced in an Astro 5 reading UI. Shared Summarise+Save sub-workflow keeps it DRY. End-to-end in one engineer-week.

n8nGeminiSupabaseAstro 5

fastapi-s07 — self-hosted AI content stack

FastAPI gateway fronting ComfyUI/Flux (image), Kokoro TTS (voice), Mermaid CLI (diagrams), Podcastfy (multi-voice podcast), and PaddleOCR — all on one OCI ARM box, called by n8n via webhooks. Replaces roughly $200/month of SaaS APIs.

FastAPIComfyUIKokoroPaddleOCROCI ARMDocker

BUD — operator cockpit on Cloudflare Pages

The operator dashboard for TOS: Cloudflare Pages + CF Access portal with eight modules — resource registry (121 items), sprint tracker, regression monitor, mobile-operator manual, weekly ops summaries, media library. Reads Supabase + Postgres; SSO-gated.

Next.jsCloudflare PagesCF AccessSupabasePostgres
03

Stack & doctrine

AI & ML
Claude Claude
Gemini Gemini
Qwen / Ollama Qwen / Ollama
pgvector pgvector
Web & app
Next.js Next.js
Astro Astro
React React
TypeScript TypeScript
Cloudflare Cloudflare
Clerk Clerk
Backend & data
Python Python
FastAPI FastAPI
PostgreSQL PostgreSQL
Supabase Supabase
Docker Docker
n8n n8n
Infrastructure
Oracle Cloud Oracle Cloud
Cloudflare Cloudflare
Tailscale Tailscale
Caddy Caddy
Linux Linux
Observability
Uptime Kuma Uptime Kuma
Prometheus Prometheus
Grafana Grafana
ntfy ntfy
Doctrines I codify and publish

thin-host architecture · resumable-pipeline pattern · single-shared-Chrome browser automation · dual-vault secrets · doctrine-as-code. Build logs and architectural breakdowns published at trendmedia.au/trendai.

04

Background

15 years across enterprise (Australian Taxation Office), offshore delivery leadership (Australia↔Vietnam), independent IT consulting, and operator roles in two of my own companies. Now consulting on agentic infrastructure and shipping production AI apps end-to-end.

Education: La Trobe University — Graduate Certificate in AI (2024–25). Swinburne University of Technology — BEng, Telecommunications and Internet Technologies (2000–05). Languages: English · Vietnamese.

05

Engagement

Best for project conversations, advisory work, or anything that needs ~30 minutes of focused thought.

Book a call
cal.trendai.au/trendai/discovery
15-min AI Readiness Scan