Mark William's Work | ContraWork by Mark William
Mark William

Mark William

AI Automation Engineer | Building Fault-Tolerant Systems

New to Contra

Mark is ready for their next project!

Cover image for Asynchronous RAG Support Architecture (n8n
Asynchronous RAG Support Architecture (n8n + Webhooks) Most AI customer support integrations fail because platforms like Shopify and Zendesk enforce strict 3-second webhook timeouts, while LLMs often take 5-10+ seconds to generate high-quality responses. To solve this, I designed a decoupled, asynchronous webhook architecture in n8n. The parent workflow catches the webhook and immediately returns a success response, while delegating the high-latency AI processing to a background sub-workflow. Inside the sub-workflow, I injected semantic context into the LLM prompts (simulating a Vector Database lookup) to constrain the AI, preventing hallucinations and securely drafting autonomous customer support replies.
0
6
Cover image for Autonomous B2B RevOps Scoring Engine
Autonomous B2B RevOps Scoring Engine (N8N + Gemini 2.5 Flash) I architected a multi-step logic pipeline to automate the triage and qualification of inbound B2B lead data. By integrating Gemini 2.5 Flash API, the system parses unstructured data to evaluate companies against a specific Ideal Customer Profile (ICP) in real-time. To ensure the sales team never misses a high-value target, I implemented dynamic Switch node routing to bypass manual qualification bottlenecks and trigger instant Slack alerts for "Tier 1" prospects. This eliminates manual processing time and ensures 100% lead coverage.
0
10
Cover image for  Fault-Tolerant Data Migration Engine (n8n
 Fault-Tolerant Data Migration Engine (n8n + REST APIs) Most data migrations fail because they aren't built to handle external API rate limits. I architected this scalable, recursive batch-processing system to migrate heavy payloads in 50-record increments while maintaining 100% stability. Key Points: - Built self-healing asynchronous error branches to catch and handle HTTP 429 (Too Many Requests) failures. - 0% data loss during high-stress migrations through automated external error logging. - Optimized throughput using deterministic loop logic to bypass integration bottlenecks.
0
16
Cover image for Enterprise KYC Vision AI Pipeline
Enterprise KYC Vision AI Pipeline
0
33