Austin Opia | Senior Laravel Engineer & Technical Lead

Overview

About the project

A product I designed and built end to end. Tenants upload knowledge (files, FAQs, documents, public repos, crawled pages); the platform embeds it, retrieves the most relevant entries for each question, and answers only from that evidence — with citations, confidence, and a human handoff when the answer isn’t there. Questions the agent couldn’t answer land in a gap tracker so the tenant can fill them and close the loop. One knowledge base serves every channel. Multi-provider LLM routing, usage quotas, Stripe/Cashier and Shopify billing, and ~250 automated test files.

Engineering

Challenges & solutions

Challenges

01Stopping the agent from inventing answers when the knowledge base has a gap
02Serving one tenant knowledge base consistently across five very different channels
03Keeping per-tenant LLM cost and rate limits under control across several providers
04Knowing which answers are going stale as tenants’ businesses change

Solutions

01Embedding retrieval with similarity thresholds, evidence assembly and a grounding check — below threshold the agent hands off instead of guessing
02Channel adapters (widget, Discord, Telegram, email, Shopify) on top of a shared tenant RAG service and conversation context
03Provider manager with rate-limit tracking, usage logging and hard chat quotas per plan
04Unanswered-query gap tracker plus staleness checks, so misses feed back into the knowledge base

All Projects Work with me on something like this

AIDEVELOPIA — AI Knowledge Layer for Businesses

About the project

Challenges & solutions