UNSENT AI

Inspiration

Communication is the most significant bottleneck in the human experience. We generate high-fidelity emotional data at a rate far higher than our social protocols allow us to transmit. The result is "Internal Latency"—a buildup of unspoken words that creates psychological drag.

Most social platforms are built to maximize retention and noise. We looked at this problem from first principles: What if we built a machine for release? We wanted to create a digital "Grounding Rod"—a sovereign sanctuary where users could offload emotional state without the risk of social repercussions or data extraction.

What it does

UNSENT AI is a multimodal atmospheric sanctuary designed for "The Unsent." It serves as a high-fidelity buffer between raw emotion and the void.

Atmospheric Orchestration: The UI is alive. Using real-time sentiment analysis, the application reconfigures its color palette, glow intensity, and background "Aura" to resonate with the user's current frequency (Angry, Sad, Hopeful, etc.).

The Smoke Ritual: A custom physics-based animation that fragments messages into character-level particles, simulating the physical evaporation of weight.

Response Modes: Users can choose their "Echo"—Silence (pure release), Closure (AI-driven kindness), or Alternate Reality (poetic "what-ifs").

Sovereign Vault: A zero-trace, local-first architecture where data is stored in the user's browser vault (IndexedDB), ensuring total privacy and data sovereignty.

How we built it

We optimized for "Fluidity-to-Latency" ratios. The stack was chosen to ensure the UI feels like an extension of the nervous system:

Logic: React 19 for concurrent rendering and TypeScript for a rock-solid emotional schema.

The Intelligence Layer: Google Gemini 3 (Flash-Preview) serves as the "UI Orchestrator." It doesn't just generate text; it returns a structured JSON "Atmosphere Packet" (Sentiment, Color, Intensity).

Physics Engine: Framer Motion was pushed to its limits to handle character-level particle dissipation and rhythmic "breathing" UI cycles.

Persistence: Dexie.js (IndexedDB) for the "Sovereign Archive," allowing for heavy data loads on the client side with zero server-side leakage.

Multimodal: Integrated the Web Speech API to allow users to "Whisper" their messages, turning spoken frequency into digital echoes.

Challenges we ran into

The Latency-Empathy Gap: Real-time sentiment analysis during typing can be computationally expensive. We implemented a debounce-driven "Aura Sync" to ensure the UI color shift felt like a natural pulse rather than a jagged update.

Particle Density: Animating 500+ characters as individual DOM elements for the "Smoke Ritual" can crush the main thread. We optimized the vector math and used hardware-accelerated transforms to keep the dissipation "butter-smooth" at 60fps.

AI Character Integrity: Ensuring the Gemini model didn't fall into "Generic Assistant" mode required a rigorous system instruction to keep its responses poetic, brief, and "Void-like."

Accomplishments that we're proud of

The Sentiment-UI Loop: We successfully closed the loop between machine learning and CSS variables. The UI feels genuinely "empathic." Zero-Trace Guarantee: We built a product that handles the most sensitive human data imaginable while having zero access to it ourselves. That is engineering ethics in practice. The "Smoke" Physics: Creating an animation that actually feels like a release of weight. Users have reported genuine physiological relief after watching their words dissolve.

What we learned

Design as Therapy: We learned that minimalism isn't just an aesthetic; it's a cognitive requirement for emotional processing. Every pixel removed increased the user's "mental bandwidth" for honesty. API as a Designer: We discovered that LLMs are better at "Atmospheric Design" than static themes. Letting the AI decide the specific hex code and blur radius of the room based on the user's pain is a new frontier for UX.