Knowy AI

Knowy AI turns scattered company data into reusable, source-backed intelligence—answering questions, remembering insights, and refreshing them once sources change.

Comment

Inspiration

As organizations rapidly adopt AI, one challenge consistently emerges: company knowledge is fragmented across Slack, Google Drive, GitHub, Notion, and countless other tools. While modern AI assistants can search these systems, they repeatedly perform the same retrieval and reasoning every time a similar question is asked. Valuable knowledge disappears at the end of each conversation.

We were inspired by the vision of enterprise intelligence platforms like Palantir, but also recognized that these solutions are often complex and inaccessible for most organizations.

Knowy AI explores this gap by introducing a lightweight Company Intelligence Layer that transforms scattered enterprise knowledge into reusable, source-backed intelligence.

What it does

Knowy AI connects to common enterprise knowledge sources and retrieves only the information relevant to a user's question.

Instead of simply generating an answer, Knowy AI creates a reusable Intelligence Object that contains:

  • A concise summary
  • Supporting evidence
  • Source references
  • Confidence score
  • Last verified timestamp
  • Freshness status

When someone asks a similar question later, Knowy AI reuses the existing intelligence instead of repeating the entire retrieval process. If one of the underlying sources changes, Knowy AI detects the change, refreshes the intelligence, and keeps company knowledge current.

Our hackathon MVP demonstrates this workflow using GitHub, Google Drive, and Slack.

How we built it

Knowy AI is built around a simple and modular architecture.

  • Next.js - powers the frontend experience.
  • FastAPI - orchestrates retrieval and intelligence generation.
  • PostgreSQL - stores reusable Intelligence Objects and their source dependencies.
  • OpenAI GPT-5.6 - powers query understanding, evidence synthesis, structured intelligence generation, and answer creation.
  • Codex - accelerated development by helping scaffold the project architecture, generate API endpoints, implement connector interfaces, create database models, and rapidly iterate on both backend and frontend components.

Our pipeline consists of:

  1. Understand the user's question.
  2. Search existing Intelligence Objects.
  3. Retrieve only relevant information from connected sources if needed.
  4. Generate a structured Intelligence Object.
  5. Store source dependencies.
  6. Reuse or refresh intelligence on future queries.

Challenges we ran into

One of our biggest challenges was deciding what NOT to build.

Enterprise AI platforms often include knowledge graphs, ontologies, complex synchronization pipelines, and sophisticated permission systems. Building those components during a hackathon would have made the project too large.

Instead, we focused on one core innovation: treating AI-generated conclusions as reusable intelligence rather than disposable chat responses.

Another challenge was designing a generic data model that could represent information from multiple enterprise systems while remaining simple enough to implement within a weekend.

Accomplishments that we're proud of

We're proud that Knowy AI goes beyond a traditional RAG chatbot.

Our MVP demonstrates that AI can:

  • Connect multiple enterprise knowledge sources.
  • Retrieve only relevant information instead of indexing everything.
  • Generate source-backed Intelligence Objects.
  • Reuse previous reasoning instead of repeating expensive searches.
  • Detect outdated intelligence and refresh it when source data changes.

Most importantly, we built a working prototype that illustrates a new way of thinking about enterprise AI: transforming company knowledge into reusable organizational intelligence.

What we learned

Building Knowy AI taught us that the biggest opportunity in enterprise AI isn't necessarily building better chat interfaces—it's building better organizational memory.

We also learned that keeping the scope intentionally small led to a clearer product. Rather than trying to replicate a full enterprise platform, we focused on demonstrating a single compelling concept that can scale over time.

Finally, Codex and GPT-5.6 significantly accelerated our development workflow. Codex helped us move quickly from architecture to implementation, while GPT-5.6 enabled reliable planning, structured outputs, and intelligence generation.

What's next for Knowy AI

This hackathon MVP is only the foundation.

Our next steps include:

  • Supporting additional enterprise platforms such as Microsoft 365, SharePoint, Jira, Salesforce, and Confluence.
  • Adding enterprise authentication and permission-aware retrieval.
  • Building richer relationships between Intelligence Objects to form a lightweight Company Knowledge Graph.
  • Introducing event-driven updates so intelligence refreshes automatically when source systems change.
  • Enabling AI agents and employees to collaborate using the same shared organizational intelligence instead of isolated conversations.

Our long-term vision is to make Knowy AI the "Company Intelligence Layer"—a lightweight platform that helps organizations transform fragmented knowledge into reusable, continuously evolving intelligence.

Built With

Share this project:

Updates