[ 01 ] Selected work

Proof, not promises.

Six systems in production, built from the Philippines for clients in the US, Australia, and Southeast Asia. Each one is described here the way we describe it internally — by what it does, what enforces it, and what it replaced. These are AI case studies, not testimonials.

15+ production systems live
300+ automated tests across platforms
104 migrations on one real-time platform
<0.8s voice-agent response latency

[ 01 ] PropTech venture (US) — Flagship

Books that reconcile to zero.

A US PropTech venture needed one platform to carry residential, commercial, and HOA portfolios — with books a CPA could trust. We built the full multi-tenant SaaS, including a complete double-entry accounting engine: rent posting, late fees, owner distributions, and trust accounting that must reconcile to zero. Three developers built in parallel under strict domain ownership.

Tenant isolation
39 tables under forced row-level security
Ledger integrity
Immutable journal entries, enforced by database triggers
Money math
Integer cents throughout. No floating point, ever.
Reconciliation
Bank-statement import via CSV and OFX
Quality gate
130+ automated tests gating CI
An accounting core a CPA can audit — built in months, not years.
  • Multi-tenant SaaS
  • Fintech-grade accounting
  • NestJS
  • PostgreSQL

[ 02 ] Advertising agency (Southeast Asia)

Every draw, provable.

An advertising agency in Southeast Asia needed promotional raffles that regulators would accept. We built a white-label platform: shoppers scan a QR code, upload a receipt, and a vision LLM validates it — with automatic model fallback when the primary degrades. Draws are atomic, RNG seeds are recorded, and every event lands in a hash-chained audit log.

Receipt validation
Vision-LLM pipeline with automatic model fallback
Audit trail
Hash-chained event log; tampering breaks the chain
Draw integrity
Atomic draws with recorded RNG seeds
Load tested
50 requests/second, p95 under 2s
Theming
Multi-brand campaigns from a single deploy
Regulator-ready raffles: every entry verified, every draw provable.
  • Next.js
  • Vision AI
  • Multi-tenant
  • Audit integrity
preview — entry

[ 03 ] Distributed staffing operation

Four tools, one quiet hub.

A staffing operation spread across time zones ran its day in four disconnected tools. We replaced them with one real-time hub: tasks, kanban, project hubs, chat, and notifications. The system tracks presence and sends an email digest only when you are actually offline — so the people online stay in flow, and the people asleep wake up informed.

Schema history
104 database migrations on one platform
Prioritization
Live priority recalculation as work changes
Access control
Staff / manager / admin roles enforced in the database
Notifications
Presence-aware digests; email only when offline
Collaboration
Threads, mentions, reactions, recurring task automation
One hub replaced four tools — and it knows when to stay quiet.
  • React
  • Realtime
  • RBAC
  • Postgres RLS
preview — hub

[ 04 ] Regional property-management operators (US)

The 2 a.m. call, answered.

Regional property operators were losing after-hours calls to voicemail — including the ones that flood basements. Our phone agents answer every call, classify emergency versus routine, dispatch urgent maintenance, and log everything else as a structured ticket. English and Spanish, switched mid-call the moment the resident switches.

Response latency
Sub-800ms over telephony-grade audio
Languages
English / Spanish, mid-call switching
Escalation
Dedicated senior-resident escalation paths
Compliance
Consent capture, DNC scrubbing, state call rules on outbound
Integration
CRM-connected booking and ticket creation
Every 2 a.m. call answered. Emergencies dispatched, the rest becomes a clean ticket by morning.
  • Voice AI
  • Telephony
  • Compliance
  • CRM integration

[ 05 ] Remote-staffing group

Three vendors. One engine.

A remote-staffing group sold three things through three vendors: candidate supply, training, and placement. We built one product — a job marketplace where employers hire certified remote operators, a training academy where candidates certify, and an AI team-builder that generates org charts and cost models on demand. Real-time messaging connects all three sides.

API surface
27+ API routers across marketplace, academy, and billing
Roles
Six-level role hierarchy
Billing
Stripe subscriptions plus placement billing
Certification
Assessment engine with five assessment types
AI tooling
Generated org charts and team cost models
Supply, training, and demand — one engine instead of three vendors.
  • Marketplace
  • LMS
  • Stripe
  • tRPC

[ 06 ] AutonoIQ in-house

A publication that runs itself.

Our own product — and the proof is public. The engine gathers industry news from 12 sources, dedupes by similarity, drafts long-form articles with an LLM, generates hero images behind quality gates, applies SEO and GEO structured data, publishes, and schedules social posts. Zero human touch. It writes the Insights section of this very site, on a daily and weekly schedule.

Sourcing
12-source news fan-out with similarity dedupe
Drafting
3-attempt LLM retry loop with corrective feedback
Image gates
Resolution, variance, and text-detection checks
Structured data
JSON-LD (BlogPosting + FAQ) on every post
Cadence
Daily and weekly schedules, fully autonomous
The blog you're reading is written, illustrated, and shipped by this system.
  • Python
  • LLM pipeline
  • SEO/GEO
  • Image generation

[ 07 ] Index

Additional engagements.

Smaller in scope, identical in discipline. A sample of other systems we have designed, built, or operate.

  • Landlord SaaS, mobile + web Southeast Asian landlord platform: iOS/Android/web, card and e-wallet payments, calendar sync with major booking channels. Expo/React Native · Payments
  • AI Document Pipeline CV anonymization and rebranding: LLM extraction to branded DOCX/PDF, PII stripped automatically. Document AI
  • Sales-Call Intelligence European B2B sales org: transcript analysis, coaching reports, deal-health scoring with a RAG knowledge base. LLM analysis · RAG
  • AI-Guided Project Intake Conversational intake that interviews prospects and outputs an engineering brief — it powers our own “Start a project” flow. LLM UX
  • Lead Extraction Platform Credit-metered B2B lead extraction with multi-locale UI and CSV export. Next.js · Data ops
  • Analytics Automation GA4 and CRM pipelines into live reporting tables with stuck-deal flagging. Workflow automation

[ 08 ] Your system

The next case study could be yours.

Tell us what your team does by hand. We will tell you what a machine should be doing instead — and what it costs to keep waiting.

Start a project Run the ROI numbers