Influencer Intelligence Service (Genkit AI + Firebase Functions)
An influencer-marketing media-intelligence platform
Overview
The AI service dedicated to influencer intelligence, built on Firebase Genkit with custom model plugins, OpenAI and DeepSeek integrations, and an evaluation harness, backed by PostgreSQL and Meilisearch.
The Challenge
Turning raw media and social data into structured influencer insight requires orchestrated LLM flows, the flexibility to swap or combine models, and a way to measure output quality. It also needs to feed results into a searchable store.
What We Built
A functions/ service organised around genkit.ts flow definitions, a genkit-functions/ set of AI flows, a custom deepseekPlugin/, reusable prompts/, and scripts/. It uses genkitx-openai for OpenAI, @genkit-ai/evaluator for quality evaluation, @genkit-ai/firebase for deployment, TypeORM over PostgreSQL for persistence, and Meilisearch for search indexing.
Technologies & Approach
Firebase Genkit provides a typed, pluggable framework for AI flows and evaluation; a custom DeepSeek plugin adds an alternative model; OpenAI handles primary inference. TypeORM + PostgreSQL persist structured outputs and Meilisearch makes them searchable.
Outcome / Impact
Established a maintainable, evaluable AI layer for influencer analytics, with the plugin architecture and evaluators enabling safe model iterative development in production flows.
Capabilities Demonstrated
- Building production LLM flows with Firebase Genkit
- Writing custom model plugins (DeepSeek) and multi-provider inference
- Adding evaluation harnesses to AI pipelines
- Persisting and search-indexing AI-enriched data (PostgreSQL + Meilisearch)