Intelligent FAQ Chatbot Migration to Serverless

Overview

Migrated an existing FAQ chatbot’s entire matching engine off constrained legacy CMS hosting onto AWS Lambda, replacing what had been running inside a content-management platform that couldn’t support the required Python NLP dependencies.

Technical Approach

• Legacy Constraint: The chatbot lived on a CMS server not built to run a Python NLP stack (tokenization, spell-correction, fuzzy-matching libraries), so the entire matching engine was moved off it entirely rather than patched in place
• Serverless Matching Engine: A single AWS Lambda Function URL now handles the full request pipeline — Thai text normalization, tokenization, synonym expansion, spell correction, and fuzzy-match scoring against a curated knowledge base
• Confidence-Based Routing: Matches route to a direct answer, a topic overview, a disambiguation prompt, or a scoped fallback depending on match confidence, rather than a single fixed response path
• Cold-Start Optimization: Expensive index-building work happens lazily on first use per warm container instead of on every cold start

Key Features

• Serverless, Dependency-Isolated: No NLP dependencies run on the original CMS server anymore
• Guardrails: Filters out-of-scope and low-signal queries before they reach the matching engine
• Config-Driven Tuning: Matching thresholds and response templates are adjustable without redeploying code

Technologies Used