I build production-grade automation systems that work.

Problem

Field service businesses struggle with operational blind spots: technician locations unknown, job status unclear, resource allocation inefficient. Managers rely on phone calls and spreadsheets to understand what is happening in the field, leading to delayed decisions and missed opportunities for intervention.

System Overview

Built a deterministic AI reasoning system that processes field service data streams to generate actionable operational intelligence. The system combines structured data preprocessing with LLM-based pattern detection to surface issues, predict delays, and recommend resource adjustments in real-time.

Core n8n orchestration workflow showing deterministic preprocessing, AI reasoning, and session persistence.

Architecture & Key Logic

Master workflow runs on configurable schedule, pulling data from Google Sheets tracking layer containing job status, technician assignments, timestamps, and completion metrics. Preprocessing stage normalizes data types, handles missing fields, and computes derived metrics such as time-since-assignment and expected-vs-actual duration deltas.

Preprocessed data is structured into a deterministic prompt template and sent to OpenAI API with temperature set to zero for consistent reasoning. LLM analyzes patterns across active jobs, identifies delayed tasks, detects resource bottlenecks, and generates specific recommendations with confidence scores. System validates LLM output structure before acceptance.

Generated insights are written to session-specific records in Firestore with timestamps and input data snapshots for auditability. Notification layer formats findings and delivers via SMS to relevant managers using Twilio. Error handling includes API retry logic, malformed response detection, and alerting for pipeline failures.

Tools & Integrations

Outcome

Production system running on scheduled intervals, processing real operational data and delivering actionable insights to field managers. Operational visibility improved through automated pattern detection, response time to emerging issues reduced, and resource allocation decisions informed by structured analysis rather than intuition.

AI Receptionist & Call Handling System

Problem

Service businesses lose revenue from missed calls and require staff to handle routine appointment scheduling. Each missed call or manual booking creates friction and potential lost revenue.

System Overview

Built a two-webhook n8n architecture that handles incoming Twilio voice calls with AI-powered conversation. The system maintains conversation memory keyed by call ID, generates contextual responses, processes appointment booking requests, and handles graceful termination.

Architecture & Key Logic

First webhook receives incoming call and initiates conversation. Conversation state is stored in Firestore using the unique call ID as the key. Each user utterance triggers the second webhook which retrieves conversation history, sends context to LLM for response generation, updates stored conversation state, and returns TwiML instructions to Twilio.

Booking logic validates requested time slots against Google Calendar availability, creates calendar events with customer details, and confirms booking to caller. Fallback handling detects prolonged silence or ambiguous requests and offers human transfer. System tracks call duration and terminates cleanly after resolution or timeout.

Tools & Integrations

Outcome

Production system handling actual customer calls with conversation continuity, successful appointment booking, and intelligent fallback to human operators when needed. Reduced missed call rate and eliminated manual scheduling workload.

System is deployed and under active development, with ongoing refinement of edge-case handling and conversational reliability.

Automated Quote & Missed-Call Follow-Up System

Problem

Service contractors send quotes and miss calls throughout the day but lack systematic follow-up. Potential customers go uncontacted, quotes expire without response, and revenue is lost to competitors with better follow-up discipline.

System Overview

Designed an n8n workflow triggered by CRM events for missed calls or sent quotes. System executes time-delayed SMS sequences, tracks customer engagement state to prevent duplicate messaging, branches logic based on customer responses, and escalates to human review after sequence completion or customer reply.

Architecture & Key Logic

Trigger nodes listen for missed call webhooks and quote-sent events from business systems. Initial validation checks if customer is already in an active sequence to prevent duplicate follow-ups. State tracking table in Firestore records sequence start time, message count, and customer response status.

First SMS sends immediately with business introduction and callback offer. Second message scheduled for 24 hours later if no response. Third message at 72 hours offers alternative contact methods. Each message send updates state and checks for customer reply via Twilio incoming message webhook.

Response detection halts automated sequence and creates task for sales team. Non-responders after final message are flagged for manual outreach. All timing and message content configurable per business needs.

Tools & Integrations

Outcome

Eliminated manual follow-up tracking and ensured every missed call and quote receives consistent outreach. Recovered lost opportunities by maintaining engagement during the critical first 72 hours. System prevents awkward duplicate messages while maximizing connection attempts.

Lead Enrichment & Outreach Automation

Problem

Manual lead processing from Apollo to email outreach platform is time-consuming and error-prone. Data inconsistencies, missing personalization, and formatting errors reduce campaign effectiveness and waste hours of manual work daily.

System Overview

Built end-to-end pipeline that imports leads from Apollo API, normalizes and validates data through Google Sheets processing layer, generates personalized outreach lines using LLM analysis, and delivers clean formatted records to Instantly for campaign execution.

Architecture & Key Logic

Daily scheduled trigger pulls new leads from Apollo with filtering criteria. Raw data written to Google Sheets staging area where custom formulas validate email formats, phone numbers, and required fields. Validation layer flags incomplete records for manual review.

Clean records pass to enrichment phase where company websites are scraped for context and sent to LLM with prompt template. LLM generates personalized opening lines based on company details, industry, and recent news. Generated content quality-checked for length and relevance before acceptance.

Final formatting step maps enriched data to Instantly's required field structure, handles edge cases like missing company names, and batches records for API submission. Error handling includes retry logic for API failures, dead letter queue for problematic records, and notification system for pipeline issues.

Tools & Integrations

Outcome

Fully automated lead processing pipeline handling hundreds of records daily with zero manual intervention. Data quality improved through systematic validation, outreach personalization increased engagement rates, and team time redirected from data entry to strategic work.