Applied AI & Workflow Automation

Automating Purchase Order Processing with an AI Agent

Context

Within customer success operations, purchase order processing required manual email triage, document extraction, standardized file naming, and organized storage. This repetitive workflow created unnecessary friction and delayed downstream execution.

Problem

Purchase orders arrived via email in inconsistent formats and required manual review, renaming, and filing. The process was time-intensive, prone to inconsistency, and slowed internal coordination.

Opportunity

Design a production-ready AI workflow that could:

  • Detect relevant purchase order emails

  • Extract key information

  • Apply standardized naming conventions

  • Automatically file documents into structured cloud storage

  • Draft response confirmations when appropriate

Solution Architecture

Using n8n as an orchestration layer, I built a multi-step AI agent integrating:

  • Gmail API for inbox monitoring and retrieval

  • Anthropic LLM for content parsing and conditional decision logic

  • Google Drive API for structured document storage

The workflow includes:

  • Conditional branching logic for PO detection

  • LLM-based extraction of relevant metadata

  • Automated document renaming and folder routing

  • Draft email response generation

  • Error handling pathways for ambiguous inputs

The agent runs in production and supports daily operational workflows.

Product Capabilities Demonstrated

  • AI workflow orchestration using n8n

  • Multi-step conditional logic design

  • API integration across Gmail, Anthropic, and Google Drive

  • LLM prompt engineering for structured extraction

  • Production deployment of automation systems

  • Converting manual processes into scalable internal products

Impact

  • Reduced manual email triage time

  • Standardized file naming and storage conventions

  • Accelerated internal response cycles

  • Clarified undocumented process assumptions through structured agent design

In documenting and building the workflow, I translated implicit team behavior into explicit system logic, improving overall process clarity.

Customer-Facing Automated Quoting Portal

Context

In a drone inspection services business, generating quotes for enterprise clients required manual back-and-forth: a client would reach out, a customer success representative would assess the scope, build a quote based on static regional pricing parameters, send it to the client, and simultaneously coordinate internally with the operations team to confirm availability before the project could be green-lit. The entire process depended on a single human touchpoint and introduced unnecessary delays on both sides.

Problem

The quoting process was bottlenecked by manual coordination. Clients had to wait for a human to interpret their request, apply pricing logic, and return a number, a process that could take hours or days depending on availability. Meanwhile, the CS representative had to run a parallel track of internal ops coordination before the project could move forward. The critical path was longer than the underlying complexity of the work justified.

Opportunity

Design a client-facing self-service quoting portal that could guide clients through a structured scoping flow, apply static regional and service-type pricing logic automatically, deliver a real-time quote without human intervention, notify the internal team via Slack to initiate ops availability checks in parallel, and remove the CS representative from the initial quoting loop entirely, compressing the critical path from days to minutes.

Solution Architecture (Current and In Development)

The initial prototype was built as a Google Form feeding into a Google Sheet, with a Google Doc adapted as a quote template. This validated the pricing logic and scoping flow but lacked real-time delivery, auth security, and Slack integration.

The production build is being developed using Claude and Claude Code for application logic and AI-assisted development, Cursor for accelerated development iteration, Vercel for deployment and hosting, Slack API for real-time internal notification to the operations team, and Google Sheets as a structured data store for quote records and audit trail.

The client flow includes three initial service type selections to determine the inspection category, site and scope detail inputs, regional and requirement-based pricing parameter selection, real-time quote generation based on static pricing logic, automated Slack notification to the operations team triggering parallel availability confirmation, and quote record logging for audit trail and follow-up. Auth security and access controls are in the roadmap for the next development phase.

Tools Used

Claude, Claude Code, Cursor, Vercel, Google Forms (prototype), Google Sheets, Google Docs, Slack API, Manus

Product Capabilities Demonstrated

  • End-to-end product ownership from manual process through prototype to production build

  • User flow design for a multi-step client-facing scoping and quoting experience

  • Pricing logic design and parameterization across regional and service-type variables

  • API integration design across Slack, Google Sheets, and hosting infrastructure

  • Identification and elimination of manual bottlenecks in a client-facing workflow

  • Iterative product development from low-fidelity prototype toward production deployment

Impact

  • Removes CS representative from the initial quoting loop entirely

  • Compresses client quote turnaround from hours or days to real-time

  • Enables parallel internal ops coordination via Slack notification at the moment of quote generation

  • Standardizes pricing application across all client requests, eliminating inconsistency

  • Creates an auditable quote record for every request

  • Shortens the critical path to project green-light by decoupling quoting from human availability