AI Agent Solutions & Business Automation
Practical LLM-powered solutions — AI-OCR order intake, reply drafting for support tickets, a daily research agent, and AI-driven product concierge
A Daily AI Research Agent — Proposal & Approval Workflow
An AI agent investigates your data every morning and registers improvement proposals. Humans approve before anything runs
Scheduling an AI Agent to Run Every Morning
Combining a local PC and the cloud to run an AI agent headlessly on a fixed schedule
Designing Research Prompts Across Multiple Data Sources
How to combine analytics, search data, sales, and CRM into a single investigation — prompts and data handoff
Human-in-the-Loop: Proposal → Approval → Execution
An approval flow where AI proposals run only after human review, plus asynchronous Q&A via Slack
AI Draft Replies for Support Tickets — Automation Where Humans Send
AI generates reply drafts for CRM support tickets. Humans always review and send — safe, realistic CS automation
Injecting Customer Context — Purchase History and Past Interactions
Feeding EC order history, past tickets, and product data into prompts to prevent off-target drafts
Classifying Inquiry Intent and Switching Response Patterns
Routing returns, shipping, and product questions to different templates and service patterns
Designing the Approval Flow and Operator Experience
Attaching AI drafts as CRM notes and the operational flow from human review to sending
AI-OCR for Fax Orders — Turning Paper Orders into Data
AI reads fax order sheets arriving by email and registers them as order data. A realistic bridge between paper culture and automation
Designing the Intake Pipeline from Email Inbox
Polling the inbox, fetching PDFs, and processing on a schedule — the foundation of automated intake
Structured Data Extraction from Forms with LLMs
Extracting item codes, quantities, and clients from inconsistent fax layouts — prompt and validation design
Per-Client Routing Rules and Misread Recovery
Managing which client's fax goes where, and the reprocessing flow when AI misreads
AI Product Concierge — Interactive Diagnosis & Recommendations
Bringing in-store service to EC: an AI that interviews customers and makes grounded product recommendations
Designing the Interview Dialog and State Management
Question flows that never miss what matters, and how to manage conversation state
Product Master × RAG for Grounded Recommendations
Vector-searching manuals and product data so AI recommendations come with evidence
Visualizing Recommendations with Tables and Charts
UI design that communicates "why this product" through comparison tables and radar charts
AI Newsletter Creation — Generating Email-Safe HTML
AI drafts the newsletter and generates HTML that won't break. Humans just review and send
Why Email HTML Is Special — Constraints for Unbreakable Layouts
Rendering differences across email clients and why table layouts and inline CSS are required
Getting AI to Write Email HTML That Doesn't Break
Enforcing constraints through prompts and validating generated output before use
Product Data Integration and a Review-First Workflow
Accurately inserting prices, images, and links from the product master, with human review before sending
Other Categories
Headless EC with Shopify
Shopify Storefront API, Next.js ISR, member pricing - fast and flexible headless EC site design
HubSpot CRM/CMS Setup & External Integration
From HubSpot implementation to custom API integrations beyond standard connectors and external tool integrations
POS & Omnichannel Integration
Unifying online and offline customer data for seamless customer experience
Accounting & Invoice Automation
Auto-generate invoices from spreadsheets with idempotency control and retry mechanisms
Order Management Integration
Bidirectional sync with order management systems and secure OAuth-based API integration
AI-Powered Site Development & Deployment
Non-engineer-friendly workflow for publishing and managing sites using Claude Code × GitHub × Vercel
Other Solutions & Ideas
Business improvement and problem-solving ideas and project examples that don't fit specific categories