Kurt Ruzzel

Software Engineer

Designing intelligent systems and custom solutions. Technical, collaborative, and obsessed with building things that actually work.

Work Creations

Exploring the intersection of AI Engineering and AI Automations.

Selected Work
Scroll up or down to see the works

AutomationSystems

AI agents and n8n pipelines built to run without supervision. Open any card for the stack behind it and the flow it follows end to end.

Scroll to explore
01

AI Voice Agent

Elle

Inbound booking and outbound follow-up calls, handled end to end with no human in the loop.

02

Lead Ops / CRM

Unified Lead Capture & Deduplication Engine

Four inbound channels collapsed into one clean, deduplicated, source-tagged Salesforce record.

03

Content & Video Engine

Mona AI

Captions, graphics, and AI video generated and published across three platforms, hands-off.

04

Document Automation

AI Document Intelligence Pipeline

Invoices, contracts, and purchase orders classified and extracted within minutes of arrival.

05

AI Legal Assistance

AckAI

Real-time conversational legal assistant, built for the AWS Cloud Club PH Innovation Cup 2025.

06

Async Voice Platform

StandUp.fm

60-second AI voice updates replace the daily sync, across any time zone.

07

Meeting Automation

Meeting Intelligence AI

Raw transcripts become summaries, decisions, and owned action items minutes after a call ends.

08

Sales Research

AI Web Research Agent

Prospect enrichment in minutes, replacing 45 minutes of manual lookup per company.

AI Voice Agent

Elle

Elle interface

An end-to-end AI coaching platform that answers inbound booking calls and runs every outbound touch after them — reminders, no-show recovery, lead nurture, and upsells — without a human in the loop.

Stack

  • Retell AI
  • Twilio
  • n8n
  • Airtable
  • Google Calendar

How it flows

  1. Inbound intake

    Twilio routes the call to a Retell AI agent that qualifies the caller and books the session against live Google Calendar availability.

  2. Five parallel outbound workflows

    An n8n automation layer runs pre-session reminders 24h before, no-show follow-ups 30min after, weekly lead-nurture calls, post-session upsells, and quarterly win-back campaigns.

  3. One record per client

    Airtable acts as the unified CRM, tracking client status, session history, outreach attempts, and conversion outcomes so every workflow reads the same state.

Outcome

Weekly admin time dropped from 20+ hours to 5. The system now handles 50+ inbound leads a month, reschedules no-shows, and surfaces upsell opportunities on its own.

Lead Ops / CRM

Unified Lead Capture & Deduplication Engine

A centralized intake system that collects, normalizes, and deduplicates inbound leads from web forms, QR codes, Facebook Lead Ads, and SMS, then pushes clean source-tagged records into Salesforce in real time.

Stack

  • n8n
  • Salesforce
  • Meta Ads
  • Typeform
  • Twilio
  • Claude

How it flows

  1. Multi-channel capture

    Web forms, QR codes, Facebook Lead Ads, and inbound SMS all land in a single n8n intake, tagged with their source on arrival.

  2. Deduplication gate

    Before any record is created, n8n queries Salesforce for existing email and phone matches, merging or flagging conflicts automatically so no duplicate ever enters the pipeline.

  3. Normalization with Claude

    Inconsistent incoming data — phone formats, name casing, address fields, campaign tags — is standardized into one schema before it is written.

  4. Pipeline routing

    Each lead is routed to the correct Salesforce pipeline stage based on its source and the service requested.

Content & Video Engine

Mona AI

Mona AI interface

An end-to-end AI content pipeline orchestrated by n8n that generates, designs, and publishes both static graphics and short-form video across Instagram, TikTok, and Facebook with zero manual intervention.

Stack

  • n8n
  • Claude
  • Canva
  • Ideogram
  • Luma
  • Instagram
  • TikTok
  • Facebook

How it flows

  1. Strategy and copy

    Claude generates the content strategy and on-brand captions for every slot in the calendar.

  2. Asset generation

    Canva and Ideogram auto-produce matching graphics via API, while Luma Dream Machine renders AI video clips for Reels and TikTok-ready cuts.

  3. Multi-platform publishing

    Publishing APIs schedule and post the finished assets to Instagram, TikTok, and Facebook automatically.

Outcome

Content turnaround went from hours of manual design and posting to minutes.

Document Automation

AI Document Intelligence Pipeline

An event-driven automation that classifies and extracts data from incoming business documents — invoices, contracts, proposals, purchase orders — within minutes of arrival, eliminating manual data entry and document triage.

Stack

  • n8n
  • Claude
  • Airtable
  • Slack
  • Gmail

How it flows

  1. Event-driven intake

    The Gmail API triggers the workflow the moment a document lands, and the PDF text is extracted for the agent.

  2. Classify and extract

    A Claude agent with a schema-validated structured-output parser identifies the document type and pulls reference number, counterparty, amount, currency, key date, urgency, and a summary from unstructured text.

  3. Type-based routing

    n8n directs each record to the correct Airtable base and table and the matching Slack channel based on Claude's classification — business judgment on top of raw extraction.

AI Legal Assistance

AckAI

A real-time AI legal assistant built for the AWS Cloud Club PH Innovation Cup 2025 hackathon, pairing conversational voice with RAG-based retrieval over a legal document corpus.

Stack

  • Agora AI
  • Amazon Kendra
  • DynamoDB
  • Amazon S3
  • React
  • Express

How it flows

  1. Conversational layer

    Agora Conversational AI runs the live voice session inside a React client.

  2. RAG retrieval

    Amazon Kendra indexes the legal corpus and returns grounded passages for each question asked.

  3. Storage and API

    An Express backend on DynamoDB and S3 keeps context storage secure and document search fast without sacrificing scalability.

Async Voice Platform

StandUp.fm

An async voice standup tool that lets teams swap daily sync meetings for 60-second AI-powered voice updates that work across any time zone.

Stack

  • Next.js
  • FastAPI
  • AWS
  • Claude
  • Deepgram
  • ElevenLabs
  • PostgreSQL

How it flows

  1. Record

    A Next.js client captures the 60-second update and stores the audio in AWS S3.

  2. Transcribe

    Deepgram handles transcription through the FastAPI backend as soon as the recording lands.

  3. Understand

    Claude Sonnet does the heavy lifting on summarization, action item extraction, tagging, mentions, and sentiment analysis.

  4. Deliver

    AWS Lambda runs the background jobs, PostgreSQL persists the results, and ElevenLabs generates the team's AI voice briefing.

Meeting Automation

Meeting Intelligence AI

An event-driven automation that turns raw meeting transcripts into structured action within minutes of a call ending, eliminating manual note-taking and follow-up.

Stack

  • n8n
  • Claude
  • Notion
  • Airtable
  • Slack
  • Gmail

How it flows

  1. Transcript intake

    The workflow fires on the transcript arriving — there is no manual note-taking step to wait on.

  2. Structured extraction

    A Claude agent with a schema-validated structured-output parser returns summaries, decisions, prioritized action items, sentiment, and next-meeting details.

  3. Sentiment escalation gate

    When a meeting shows unresolved tension the workflow alerts the team in Slack automatically, adding business judgment on top of raw extraction.

  4. Distribution

    Records are written to Notion and Airtable and sent on via the Gmail API so the follow-up is already in flight.

Sales Research

AI Web Research Agent

An automated research pipeline that enriches sales prospects in minutes instead of the 45 minutes of manual lookup it replaces, closing the loop from raw company name to actionable report.

Stack

  • n8n
  • Claude
  • Tavily
  • Apify
  • Notion
  • Slack

How it flows

  1. Parallel collection

    Tavily AI web search, Apify website and Google Maps scraping, and direct HTTP scraping run side by side and merge into a single unified research context.

  2. Synthesis

    A Claude agent with a structured output parser turns the multi-source data into a validated JSON prospect profile — summary, pain points, competitors, enrichment score.

  3. Delivery

    The profile is written to a Notion database and announced in Slack, closing the loop from company name to sales-ready report.

Toolkit

Tools I build with

The stack behind the work — chosen for speed, type safety, and getting out of the way.

React
Next.js
TypeScript
JavaScript
Tailwind CSS
HTML5
CSS3
Astro
Vercel
Node.js
Express
FastAPI
ElevenLabs
Luma
Slack
Airtable
Notion
Gmail
Google Calendar
Typeform
Twilio
HubSpot
Salesforce