Creative Inference

Building at the Frontier of AI

An AI technology studio crafting intelligent agents, natural language search, and AI-powered platforms that ship to production.

View Our Work Get in Touch

Who We Are

Creative Inference is an independent AI technology studio. We design, build, and ship products powered by artificial intelligence — from autonomous agents and natural language search to AI-driven curation and content generation. Local directories go stale overnight; we build the agent pipelines, cloud infrastructure, and engagement layers that keep products fresh and ship them to production.

What We Do

Full-stack capabilities across the applied-AI stack.

AI Agents & Automation

Multi-agent systems that detect, reason, and act — automating pipelines that used to need a human in the loop.

Natural Language Search

Semantic and conversational search that understands intent, powered by large language models and embeddings.

AI Curation & Content

Automated curation, ranking, and AI-generated content that keeps products fresh without manual effort.

Cloud Infrastructure

Production-grade deployments on Google Cloud — scalable, observable, and built to run AI workloads reliably.

What We've Built

AI-powered products, shipped to production.

Under the Hood: BarDC

Happy hours change, events disappear, ratings drift — and editorial teams can't manually keep hundreds of bars current. Here's how Bar DC turns the messy open web into a nightlife guide that refreshes itself overnight.

15+ Autonomous agents
Gemini Reasoning & generation
Nightly Self-refreshing pipeline
$0 idle Serverless, scale-to-zero

The Agent Pipeline

Every night a fleet of specialized AI agents runs end to end. Each agent does one job well and hands off to the next — with a human-in-the-loop approval lane when the scrape is uncertain.

  1. 01

    Crawl

    A headless-Chrome agent visits each venue's website, walking up to 10 pages and storing the raw content.

    Puppeteer
  2. 02

    Extract

    Gemini reads the unstructured HTML and pulls out structured events, happy-hour specials, and social links.

    Gemini
  3. 03

    Enrich

    Google Places adds official ratings, hours, price levels, photos, and fresh reviews for every venue.

    Places API
  4. 04

    Curate

    More agents categorize venues against a master taxonomy and write structured, on-brand descriptions from real reviews.

    Gemini
  5. 05

    Editorialize

    A multi-factor scorer picks the night's best events, then Gemini writes the daily "Tonight" article and generates its hero image.

    Gemini + Image
  6. 06

    Publish

    Eleventy rebuilds the whole site from the fresh data and deploys to a global CDN — live before the morning rush.

    Eleventy

How It's Built

A serverless, scale-to-zero architecture on Google Cloud. It costs nothing when idle and handles traffic spikes for free.

Data

Cloud Firestore

The single source of truth — venues, crawled content, events, happy hours, news, and the daily editorial all live here.

Compute

Cloud Run + Gemini

The agent fleet runs as a containerized Cloud Run Job, triggered nightly by Cloud Scheduler, with Gemini as the reasoning engine.

Delivery

Eleventy + Firebase

A static site generated at build time and served from Firebase Hosting's CDN — fast, cheap, and resilient.

Meet "Bar Matey"

A witty AI bartender lives in the app's chat rooms. Ask it where to go and it answers from a continuously regenerated context of every venue and upcoming event — grounded in real data, aware of which neighborhood or venue room you're in. It's retrieval-augmented generation, wired straight into Firestore via a Cloud Function trigger.

A Daily Product, Not Just a Directory

Beyond the static site: Google / Apple / email accounts, favorites and “For You Tonight,” multi-axis ratings, “I’m going tonight” presence, contributor event submissions, morning push digests, a three.js night-city map, and a React admin — the engagement layer that turns structured venue data into a product people return to.

Get in Touch

Have a project in mind? Let's talk.

hello@creativeinference.com GitHub