Sentient

SENTIENT is a spatial reasoning engine designed to move AI beyond the limitations of linear text-based chat. While traditional interfaces force users to “trust the black box,” SENTIENT externalizes the model’s internal cognitive process by rendering a dynamic, collaborative knowledge graph.

Our application categorizes the agent’s output into four distinct logical layers:

Concepts (Core ideas)

Tensions (Contradictions or risks)

Evidence (Data-backed support)

Synthesis (Actionable conclusions)

Technical Execution & Frameworks:
To achieve a truly agentic experience, we utilized a sophisticated stack:

CopilotKit: Integrated to provide a canvas-aware sidebar. The agent doesn’t just suggest text; it actively manipulates the UI, adding, connecting, and refining nodes based on the ongoing conversation.

Gemini 3.1 Pro: Leveraged for its superior long-context window and reasoning capabilities, allowing the agent to ground its logic in massive datasets pulled from our MCP connectors.

Anthropic MCP (Model Context Protocol): We implemented live connectors for Notion, GitHub, and Gmail. This enables the agent to pull real-time private context—like project specifications or email threads—to ground the visual graph in reality rather than hallucination.

Generative UI: Unlike static dashboards, our UI is fully interactive. The agent renders “live” forms and status badges (Todo/Doing/Done) that sync across users in real-time via Supabase Realtime.

Our team utilized a baseline knowledge-graph visualization library we experimented with previously, but the entire Agentic Interface layer, CopilotKit integration, MCP connector logic, and Realtime Collaboration were built from scratch during this hackathon to meet the Generative UI theme.