Semantic Search
Unlike keyword-based search that relies on exact character matches, Corgtex uses semantic vectors. When you search for terms like “How do we handle APAC vendor evaluations?”, the system embeds your query and performs a nearest-neighbor vector search in the PostgreSQL database. This means the system understands intent and context, returning the relevant chunks even if the exact words “APAC” or “vendor evaluation” weren’t explicitly written in that specific combination.Ask (Conversational AI)
The “Ask” interface (available in the web UI chat and via Slack integration) sits on top of the search layer to provide direct, conversational answers.The Query Loop
- User asks a question (e.g., “Why did we exit the APAC market in 2024?”)
- Retrieval: Corgtex runs a semantic search against the Brain and retrieves the top most relevant chunks (respecting the user’s RBAC permissions).
- Synthesis: The system injects these chunks into the context window of the LLM.
- Response: The AI synthesizes the answer directly from the provided data.
- Citations: The AI appends explicit citations linking back to the exact files, rows, or articles it used to generate the answer.
Zero Hallucinations: Because the agent is strictly instructed to only answer based on the retrieved context, you eliminate the risk of the model inventing company policy.
Real-world impact
When you see your entire workforce’s knowledge:- No lost history: Dissenting opinions, risk analysis, and board decisions are accessible instantly.
- Faster onboarding: New hires can simply ask the brain contextual questions instead of waiting on busy managers.
- Cross-department visibility: If two departments are unknowingly duplicating a vendor evaluation, the agent can flag the overlapping context when you query current active projects.