2026

Aegis Agent

Autonomous portfolio concierge backend: agentic RAG over Supabase pgvector, intent routing, confidence-gated answers, Google Calendar and Meet booking, and human-in-the-loop handoff for sensitive leads. FastAPI service with LangGraph-style orchestration, session memory, and lead logging. Powers the live chat agent on this portfolio.

Python

FastAPI

LangGraph

RAG

Supabase

pgvector

Gemini

Google Calendar

On this page

Context Architecture Safety & handoff Stack Links

Context

Aegis Agent is the backend behind the conversational assistant on this portfolio: it answers questions about experience and projects, retrieves grounded context, and can escalate to human follow-up—including Google Calendar booking when it makes sense.

Architecture

FastAPI service with agentic RAG over Supabase pgvector, intent routing, and confidence-gated replies so the UI never invents facts about private timelines. Session memory and lead logging keep conversations auditable. Orchestration follows LangGraph-style patterns suitable for extending with more tools later.

Safety & handoff

Human-in-the-loop paths are first-class: when a request is sensitive or low-confidence, the design favors clear handoff over a clever but wrong answer. That bias matters for any assistant that sits on a hiring lead or a serious inquiry.

Stack

Python, FastAPI, LangGraph-oriented flows, RAG with pgvector on Supabase, Gemini for model calls, Google Calendar API for booking—not a toy demo, but a structure you can extend with evals and monitoring as traffic grows.