Not a tool that uses AI. The semantic infrastructure FeedScale, DashAI and Voxscope run on — and that external LLMs will run on via MCP. AI-native for media intelligence.
Until 2024, GeriAI was described as: "we use foundation LLMs to tag content in MySQL". Technically accurate. Strategically weak.
The paradigm changed. Models are no longer tools inside the workflow — they are the environment where the workflow happens. The useful question is no longer "what AI does GeriAI use" but "what infrastructure can LLMs run on to do media intelligence". The answer is GeriAI.
GeriAI is organised in three independent yet chained layers. Each has its own repository and role. Together they form the semantic infrastructure that powers the ecosystem.
Takes collections of news and posts stored in MySQL and turns them into structured data: sentiment, tone, per-client topic categories, brand mentions, metrics. Uses foundation LLMs with sector-calibrated prompts. Result: a persistent semantic catalogue, not ephemeral analysis.
Agents run over the catalogue without human intervention. They surface featured topics, detect emerging trends, generate content drafts, alert in real time. Delivered by Telegram, email or SLSS. Not rule-based notifications — agents deciding what's worth reporting.
REST API that interprets SAT natural-language briefings and automatically configures the whole stack for a new client: consumers, workers, analytics, mochis, brands, keywords. Service onboarding goes from hours to minutes.
The Mochis are GeriAI's hands. Each archetype has a specific mission inside the semantic catalogue, but they all share the same reasoning engine.
Automatically identify the most relevant topics of the period.
Detect emerging trends before they go viral.
Generate derived content drafts from the catalogue analysis.
Alert in real time about new relevant content appearing on the radar.
Mochis are deployed per client and sector. Each instance is trained with the specific tags and criteria of the contracted service.
Everything above runs on a proprietary tech stack. These are the disciplines we command — the ones that make the engine differentiable, not just another integration.
Natural language processing developed in-house. Not a wrapper on top of external APIs — it's the understanding logic itself.
Entity recognition and disambiguation in 45 languages. People, brands, places and products identified and sector-calibrated.
Model selection per use case and client sensitivity. Not tied to a single provider — the GeriAI architect makes the call.
Vector representation of text and meaning-based retrieval. Rank fusion (RRF) combining fuzzy and semantic search.
No-Bias convention: categories adapt to the client's sector, not generic labels. Special `sin_tag` tag as the universal escape.
MySQL + BigQuery + 4 Elasticsearch clusters. Millions of mentions per day, low latency, full queryable history.
MCP (Model Context Protocol) is the open Anthropic standard letting models like Claude, GPT or Gemini connect to external data sources without bespoke integration. An MCP server exposes resources, tools and prompts the model can use directly.
GeriAI has all the ingredients to be a first-class MCP server: per-client semantic catalogue in MySQL, tagging engine, alert generation, Mochi agents. Exposing it as MCP will let any external model query our infrastructure and execute analysis without bespoke integration — the natural generalisation of the AI-native positioning.
See the rest of our research lines →Custom services on GeriAI: data lakes, MCP integrations, specific sectors, calibrated models. Tell us what you need to measure and we build the infrastructure on top of the cognitive layer we already have.
Talk to the teamResponse in under 24h · SAT + Sales