ROUTEBusiness workflows
AI Gateway Upgrade for Marketing Platform
Roger Neel@mavenroger
This project upgrades an internal marketing platform's features, including trend signal classification, chat context routing, and branded-campaign detection, by integrating with Vercel's AI gateway and utilizing a JEV mechanism for improved semantic understanding and efficiency.
TakeThis often drifts into duplicate slugs. - With JEV: My rank rubric already reads as five ordered levels, so it maps directly to a Score question, with Choices for sentiment and urgency and one Noul per topic in a controlled taxonomy.
Time to @typesafeai JEV it up via @vercel's AI gateway for the next few hours
My internal marketing ad and newsjacking platform is about to get an upgrade across 3 features this afternoon:
Feature #1: Trend signal classification
- Current state: This is the best candidate. Right now, I send about 200 signals a day to Haiku, which parses JSON for relevance, sentiment, urgency and free-text topics. This often drifts into duplicate slugs.
- With JEV: My rank rubric already reads as five ordered levels, so it maps directly to a Score question, with Choices for sentiment and urgency and one Noul per topic in a controlled taxonomy. Only the one-sentence summary needs to stay generative.
Feature #2: Chat context routing
- Current state: It uses eight regexes to decide which data to load contextually for users looking at various reports. Are you looking at FB ads? Google ads? A newsworthy story?
- With JEV: One Noul per data need does this semantically in about 100ms.
Feature #3: Branded-campaign detection
- Current state: As anyone who's run Google SEM ads knows, branded and non-branded campaigns are totally different. Different strategy and analysis required. It was way to slow to detect with LLMs, so again using regex.
- With JEV: A Noul over campaign name, ad group names and sample keywords is more robust.
Locked in and shipping.