Think BIG

PORTFOLIO

Full-Stack
AI Agent
Architecture.

Every build shown here runs on Think BIG's own systems — our infrastructure is our proof of work.

查看中文版 →
Full-stack AI Agent architecture diagram — multi-agent orchestration from data input to coordinated output

OVERVIEW

What we build

Think BIG service overview: operations monitoring, document automation, AI customer service, local knowledge base, personal AI assistant, security gate

Enterprise AI Agent deployment

TB FRAME guides four phases — Assess, Build, Validate, Operate — each with defined deliverables and acceptance criteria, so deployment is traceable and reviewable.

Personal AI assistants

Long-running AI assistants installed on your own computer. You bring your own API key to keep costs predictable and avoid vendor lock-in. Think BIG engineers handle remote setup and maintenance.

SELECTED WORKS

Nine selected builds

Multi-system operations dashboard showing real-time status, anomaly alerts, and system health indicators mockup

02

Operations Monitoring Hub

24/7automated watch
PROBLEM
Multiple systems each had their own alerts. Issues only surfaced when someone noticed and reported them — response was always delayed.
APPROACH
Built a unified monitoring dashboard that aggregates system states and pushes anomaly notifications to designated channels automatically.
RESULT
Issues are visible in real time. Response time dropped sharply. No more relying on manual polling.
  • Cloudflare Workers
  • Telegram Bot
  • Auto alerting
  • Status dashboard
Enterprise knowledge Q&A interface — AI answers from company documents with source citation mockup

03

Enterprise Knowledge Q&A

Privateknowledge retrieval
PROBLEM
Company information was scattered across platforms. Staff spent time searching, then still weren't sure they had the latest version.
APPROACH
Built a private RAG knowledge base. The AI answers directly from company documents and cites the source.
RESULT
Routine questions are answered by the AI. Staff focus on decisions that need judgment. Answers are traceable to source documents.
  • RAG
  • Vector database
  • Document indexing
  • Private deployment
Document intake automation flow — AI classifies incoming documents and generates reply drafts mockup

04

Document Automation

Repetitive workhandled automatically
PROBLEM
Receiving, classifying, and drafting replies consumed large amounts of staff time daily, leaving little room for judgment-intensive tasks.
APPROACH
AI reads incoming documents, classifies them, and generates a draft reply in the correct format. Staff review and send.
RESULT
Document throughput increased. Staff only review final drafts. Misdirected and missed replies decreased significantly.
  • Document parsing
  • Auto classification
  • Draft generation
  • Human review gate
Website AI customer service interface showing conversation flow and escalation mechanism mockup

05

Website AI Customer Service

24hinstant response
PROBLEM
Enquiries outside office hours went unanswered. Leads were lost. Customers waited until the next business day.
APPROACH
Deployed an AI support agent trained on the knowledge base. It handles common questions and escalates complex cases to staff.
RESULT
Round-the-clock instant replies. Online enquiry conversion improved. Customer wait time dropped substantially.
  • Cloudflare Worker
  • Supabase
  • Knowledge base
  • Escalation logic
Emergency voice alert system — trigger conditions, automatic call, and confirmation chain mockup

06

Emergency Voice Alert

Alertconfirmed to a person
PROBLEM
On-site incidents need immediate attention. Text messages are easy to miss, slowing down incident response.
APPROACH
When a trigger condition fires, the system places an automated voice call and keeps trying until the person confirms receipt.
RESULT
Zero missed incident notifications. Response time shortened. Accountability chain is logged.
  • Twilio
  • Voice call
  • Trigger automation
  • Accountability log
Local knowledge base architecture — vectorisation and retrieval run entirely on company hardware mockup

07

Local Knowledge Base

Datastays on your hardware
PROBLEM
Uploading company documents to cloud services raised data sovereignty concerns, but AI-assisted search was still needed.
APPROACH
Built a local vector database on the company's designated hardware. Vectorisation and retrieval happen entirely on-device.
RESULT
Data never leaves company hardware. Knowledge queries return results with verifiable source citations.
  • Ollama
  • Local vector store
  • Data sovereignty
  • Private deployment
Personal AI assistant management interface showing model switching, usage tracking, and custom instructions mockup

08

Personal AI Platform

Payfor what you use
PROBLEM
Individual users wanted a long-running AI assistant without being locked to one provider or having unpredictable subscription costs.
APPROACH
Built a cloud AI assistant platform. Users bring their own API key, switch between models, and pay only for actual usage.
RESULT
Subscribers control costs through model selection. Usage is transparent and trackable.
  • OpenClaw
  • Hermes
  • Multi-model
  • Bring your own API key
Pre-deployment security gate — OWASP checklist review producing a traceable audit record mockup

09

Pre-Deployment Security Gate

Every deployhas a written record
PROBLEM
Systematically verifying security risks before an AI system goes live was difficult. Review standards varied by reviewer.
APPROACH
Built a security review gate (D4). Each deployment checks against an OWASP red-line list and produces a written audit record.
RESULT
Review process is standardised. Every decision has a written basis. Traceability survives team changes.
  • OWASP
  • Deployment gate
  • Audit trail
  • Automated checks
LINE messaging automation workflow — message routing, auto-reply, and human handoff mockup

10

LINE Automation Workflow

Routine messageshandled automatically
PROBLEM
Both internal collaboration and customer communication happened in LINE. High message volume meant manual sorting and frequent missed items.
APPROACH
Integrated LINE Messaging API to route messages by type. Common questions trigger auto-replies. Complex cases go to staff.
RESULT
Routine messages handled without staff. People focus on conversations that actually need judgment. Missed-message rate dropped.
  • LINE Messaging API
  • Webhook
  • Message routing
  • Auto-reply

USE CASES

Eight entry-point scenarios

Each scenario is a self-contained starting point. Pick the most painful one, define acceptance criteria, and expand once it's working.

  • Sales follow-up & proposals — Inquiries don't go cold.
  • Finance reconciliation
  • Meeting notes & action items
  • Business data analysis
  • Quote draft generation
  • Inventory anomaly monitoring
  • Marketing copy & scheduling
  • Employee onboarding
Sales follow-up scenario — AI consolidates visit notes and produces a next-action checklist mockup
Inquiries don't go cold.
Business analysis scenario — AI aggregates multiple reports and produces a key-metrics summary mockup
Marketing copy scenario — AI generates social media post drafts in brand voice mockup

DELIVERY

TB FRAME deployment flow — four-phase milestones and deliverables: Assess, Build, Validate, Operate mockup

TB FRAME Delivery Process

Not install-and-leave. Four phases, each with defined deliverables and acceptance conditions — so every deployment is trackable and reviewable.

  1. 01Assess — Define process scope, data readiness, and maturity. Output: written assessment report.
  2. 02Build — Implement AI Agents based on the assessment. Complete configuration, integration, and initial testing.
  3. 03Validate — Test against the written acceptance criteria. Output: acceptance record.
  4. 04Operate — One year of governance and maintenance: issue tracking, maintenance checks, and usage adjustments — all logged.

RESULTS

thinkbigtw.com results snapshot — AI Agent operations, customer service, and content automation metrics mockup

Think BIG runs it first

Every build in this portfolio runs on Think BIG's own website, operations pipeline, or internal workflows. We use our own systems as the proving ground — so our knowledge of what breaks, what works, and what scales comes from direct experience.

9
Selected builds
8
Entry-point scenarios
4
TB FRAME phases
1yr
Annual operations support

Start with one thing. Make it work. Then expand.

Book a consultation See the delivery process