System StatusOnline
NAV
AnswerLyzer is an advanced SaaS platform for monitoring Brand Share of Voice in Generative AI. It allows companies to understand how they are perceived by LLMs (ChatGPT, Gemini, Perplexity). The application simulates user queries across multiple models and uses an 'LLM-as-a-Judge' architecture to automatically grade brand visibility, sentiment, and ranking.
The system sends user-defined prompts (e.g., 'Best video editing software') to multiple AI models in parallel.
We don't just store text. We feed the output back into Gemini 2.5 Flash with a strict system instruction: 'Analyze this text. Is Brand X mentioned? What is the sentiment?'
The Judge returns structured JSON data via Schema Enforcement, converting unstructured text into hard analytics.
Aggregated data is presented as Share of Voice charts and sentiment trends.
LLMs output raw text blocks. Solved by using Gemini's `responseSchema` in Edge Functions to force strictly typed JSON output.
Multi-model analysis takes time. Implemented an async queue pattern. The frontend receives a '202 Accepted' and listens for results via Supabase Realtime.
Models lack real-time data. Solved by enabling Google Search Grounding (`tools: [{ googleSearch: {} }]`) to simulate an AI Search Engine experience.
While the current iteration uses TypeScript Edge Functions, the logic is ready for a Python port suitable for Hugging Face Spaces.
Initiate protocol. Establish connection. Let's build something loud.