The AI service layer behind an AI dating concierge — conversational guidance grounded in live web search.
Alfred is an AI dating concierge. This project scope covered the AI service layer only: a production-ready FastAPI backend built around a provider-agnostic LLM abstraction, so the underlying model can be swapped without touching the product around it, plus a live SerpAPI search integration for grounding advice in current, real-world information.
Primary technologies
01 / Context
Alfred is an AI dating concierge. This project scope covered the AI service layer only: a production-ready FastAPI backend built around a provider-agnostic LLM abstraction, so the underlying model can be swapped without touching the product around it, plus a live SerpAPI search integration for grounding advice in current, real-world information.
Engineering problem
A dating concierge needs to reason conversationally about a user's specific situation, not just recite generic advice — and static model knowledge alone can't keep answers current.
Approach
Built a provider-agnostic LLM abstraction so the concierge isn't locked to a single model vendor, and integrated SerpAPI so responses can draw on live search results. A full pytest suite covers the service layer.
02 / Architecture
Layer 01
The integration surface handed to the product's backend team.
Layer 02
A single internal interface over the model vendor, so the underlying model can be swapped without touching the product around it.
Layer 03
SerpAPI integration so advice can draw on current, real-world information rather than static model knowledge.
03 / Engineering judgment
Model quality and pricing move quickly. The abstraction means a vendor change is a config change, not a rewrite.
Static model knowledge goes stale. Live retrieval keeps guidance anchored to what is actually true now.
04 / Stack
Service
AI
Quality
05 / Capabilities
Conversational dating guidance grounded in live web search
Swappable model provider behind one interface
Full pytest coverage across the service layer
06 / Challenges
Challenge 01
Problem
A concierge has to reason about a user's specific situation, not recite generic advice — and static model knowledge cannot keep answers current.
Solution
Paired a conversational LLM layer with live SerpAPI retrieval, so responses combine reasoning with current information.
07 / Outcome
Delivered as a handoff-ready AI layer with test coverage in place, for a separate backend team to integrate into the product.
Next step
I work on the AI layer — retrieval, reasoning, and the plumbing underneath — and hand it over ready to integrate.