The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
One reliability pattern I use in Revenue Ops Control: keep browser state, orchestration, and verification separate.
The setup is deliberately boring:
• GitHub is an auditable command bus.
• A local agent keeps the authenticated browser session on the machine.
• Browser actions run through Chrome DevTools Protocol and small deterministic commands.
• Each result is written back as structured execution evidence.
• A failure reruns the exact failing step instead of replaying the whole workflow.
That last part matters most. With browser automation, a full replay can create duplicates or move state you already verified. Targeted retries make debugging safer.
I use this pattern for browser/back-office automation and production debugging.
Stack: Python, Node.js, CDP, GitHub Actions.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Challenges
View all
Related posts
Project Title: Production-Grade n8n Lead Capture & Routing Engine
Project Overview:
Most marketing campaigns bleed high-ticket leads due to silent automation failures—unformatted phone numbers, API rate limits, or network timeouts. I engineered this fault-tolerant lead capture pipeline in n8n to guarantee 100% data integrity for incoming web and ad traffic.
The Architecture & Data Flow:
Production Ingress & Handshake: A custom webhook instantly catches payload submissions and holds the HTTP connection open until processing finishes, returning explicit 200 OK or 400 Bad Request signals directly to the frontend form.
Automated Data Sanitation: A javascript logic layer intercepts raw data, strips whitespace, standardizes casing, and normalizes regional phone numbers into strict E.164 international formats before it ever touches a database.
Smart Deduplication Engine: The system queries existing database records in real-time. If a match is found, it updates the engagement timestamp and routes the duplicate to an isolated audit log, preventing CRM pollution without ignoring the prospect.
Multi-Tab Database Sync: Clean traffic is routed into segmented Google Sheets or database tables (Valid Leads, Invalid Leads, Duplicates), ensuring sales reps only see verified data.
Speed-to-Lead Dispatch: Instantly triggers an internal email/Slack alert to the sales team while simultaneously firing a contextual, personalized auto-reply to the prospect.
Decoupled Error Recovery: If any external node fails, an isolated global error-trigger captures the exact n8n execution URL and stack trace, alerting the administrator instantly so no lead is ever silently dropped.
Tech Stack: n8n, JavaScript, REST APIs, Google Sheets API, Postman (for payload stress testing).
Client: live streamer (yas_garbage)
Role: solo - analysis, architecture, development, delivery
The problem. The streamer collected video links from chat and tracked them by hand, which cost air time and left the audience with no idea what was queued or how long the session would run.
What I built. A local desktop tool with two surfaces: a control panel the streamer uses off-camera, and a transparent overlay embedded in OBS as a Browser Source. Adding a link pulls the title, thumbnail and duration from the YouTube Data API; the overlay updates instantly with the new queue and the remaining total runtime.
Engineering decisions worth naming. Everything runs on the streamer's own machine - no hosting, no accounts, no viewer data leaving the computer. The server listens on localhost only, validates request origin, and keeps the API key out of every page and message it sends. The overlay is strictly read-only and authenticated by a generated token, so a leaked URL cannot modify anything. State is broadcast as versioned full snapshots rather than diffs, which makes the on-air view self-healing after any disconnection.
Delivery. Packaged as a single Windows executable with tray control and optional autostart, covered by an automated smoke suite of ~1,500 checks run against both source and the built binary.
Stack: Node.js, WebSocket, YouTube Data API v3, obs-websocket, vanilla JS/CSS front end, Node SEA packaging.
Built a self-hosted observability platform for error tracking, performance monitoring, sessions, releases, and alerts.
