Asif I.

Most AI support chatbots deflect tickets. I build the ones that resolve them. I spent seven years on Upwork's own support platform — Zendesk, Help Center, and the AI chatbots behind a support team of 300+ agents handling ~600K tickets a year — the last three as Engineering Lead. WHAT THAT WORK PRODUCED - Engineered the Forethought–Zendesk chatbot integration behind a 58% chatbot resolution rate — Zendesk published the result in their customer story - Moving off the previous chatbot took Upwork from 25–30% containment to a 52–65% average self-serve rate — per Forethought's own case study - Cut Zendesk API usage 60% by replacing a scheduled tag-sync job with real-time GraphQL data, which also ended live-chat misrouting - Built a three-way messaging model for dispute cases, so freelancer, client, and agent shared one ticket instead of three conversations (also Zendesk-featured) - Raised CS agent productivity 70% with internal Zendesk apps - Shipped eight Help Center redesign launches on a modular theme framework deployed through the Zendesk Theme API - Ran chatbot platform work end to end — Ada (2017–2021), Forethought (2021–2024) Engineering since 2012. 100% Job Success on Upwork. WHAT I BUILD - Zendesk implementation, migration, and cleanup of instances that grew messy - AI chatbots that resolve — RAG over your real knowledge base, connected to your real customer data, with clean handoff to agents - Zendesk ↔ product integrations: REST, GraphQL, webhooks, OAuth, custom apps - Help Center / Guide theme development and information architecture - Agent-productivity apps, triggers, macros, workflow automation - LLM infrastructure: RAG pipelines, MCP servers, n8n agent workflows, evaluation and observability RECENT BUILDS - Zendesk Skills — an expert knowledge pack that gives AI coding agents real Zendesk theme-development competence - Finch — a RAG support chatbot that answers from your knowledge base, opens a Zendesk ticket when it can't, and hands off to a live agent (in development) STACK JavaScript/TypeScript, Node, React, Next.js, Python, PHP/Laravel · OpenAI, Claude, Gemini, Ollama, LiteLLM, Langfuse, MCP · PostgreSQL, Supabase + pgvector, MySQL, MongoDB · Docker, Terraform, AWS, GCP HOW I WORK I find the root cause before I write code. If your chatbot plateaued or your API bill climbed, there's a reason sitting in the architecture, and patching the symptom just moves the problem somewhere else. I'll tell you what's actually wrong — including when the answer is smaller than the project you came in asking for. Send me the messy version of the problem. I'd rather see the real thing than a tidy spec.