Navigation Assistant — Multilingual AI Knowledge System
Production AI knowledge assistant delivering grounded, context-aware answers across 54 structured product guides with multilingual support, routing, validation, analytics and memory.
Business Challenge
Internal teams needed reliable answers to complex SaaS configuration, workflow and implementation questions. The previous assistant lost follow-up context, struggled with indirect and multilingual requests, and could return incomplete guidance. Routing failures and unanswered questions were also difficult to analyse.
Solution Delivered
Independently designed and built an AI-native knowledge system using Claude Sonnet, OpenAI embeddings, Node.js, TypeScript, PostgreSQL and pgvector on AWS. Structured 54 guides as metadata-rich skills and implemented multi-skill routing, dialogue classification, answer modes, grounded generation, validation and safe fallback. Added multilingual responses, session and persistent memory, semantic caching, analytics, benchmark evaluation and asynchronous jobs. Shipped the complete solution to production.
Outcomes Achieved
Created a production knowledge system supporting 54 structured guides and handling direct questions, indirect requirements, follow-ups, confirmations, corrections and multi-step requests. Improved knowledge accessibility and answer consistency while providing analytics for routing failures, fallback usage, unanswered questions and knowledge gaps.
Measurable Business Outcome
Successfully deployed to production with 54 structured guides, multilingual same-language interaction, session and persistent memory, multi-skill routing, validation, analytics and safe fallback behaviour.