GranPal

Inspiration

Every member of our team shares a common heartache: watching a loved one struggle with the isolation and memory loss that comes with aging. We’ve seen the frustration of forgotten pills and the sadness of stories fading away.
Current solutions feel like "surveillance"—cold, clinical, and intrusive. We were inspired to build something that feels like presence.
Our mission was to create a bridge between medical safety and emotional warmth, ensuring that when family can’t be there, GranPal is.

What it does

GranPal is a "Digital Granddaughter" that transforms elderly care from passive monitoring into an effortless, 2-way conversation. Using a "Magic Tablet" interface, GranPal (Chloe, the default name) uses vision and voice to:
Companionship: Engage in real-time, empathetic chat to reduce loneliness.
Contextual Support: Notice environmental cues—like an empty water glass or a misplaced pill bottle—and bring them up naturally.
Clinical Insights: Extract medical logs and cognitive trends for families and doctors without ever feeling like a "spy."

How we built it

We engineered a dual-model architecture to balance real-time empathy with deep reasoning:
Gemini 2.5 Flash: Powers the Live API for Chloe’s real-time interaction. It processes live video frames and audio to ensure Chloe can "see" and "hear" with near-zero latency.
Gemini 3 Pro: Operates as the "Cognitive Orchestrator," using its 1-million-token context window to maintain a persistent life-long memory.

Challenges we ran into

The Privacy Tightrope was our biggest hurdle. We had to ensure Grandma never felt "watched." We spent days fine-tuning the vision filtering to ensure Chloe only processes what is necessary for safety and storytelling, redacting sensitive personal moments automatically. We also battled latency; to make Chloe feel human, we had to optimize the WebSocket stream to ensure her responses remained under 500ms.

Accomplishments that we're proud of

We are incredibly proud of creating an AI that doesn't feel like "software." Seeing Chloe successfully identify a household object and weave it into a reminiscence therapy session felt like magic. We successfully balanced high-level clinical data extraction with a persona that is genuinely warm and comforting.

What we learned

We learned that AI’s greatest strength in healthcare is Dignity. We discovered that multimodal LLMs can turn "monitoring" into "reminiscence." We also learned the technical nuances of managing long-term context—keeping a conversation coherent over weeks of interaction is a massive engineering challenge that Gemini 3 Pro handled beautifully.

What's next for GranPal

The future of GranPal is about expanding the "Memory Palace." We plan to:

Wearable Integration: Moving from a tablet to AR glasses for a true "walk-along" companion.
Multi-Family Sync: Allowing multiple family members to contribute photos and voice notes to Chloe’s knowledge base.
Visualize the life-memory: Built with React and Three.js, we visualize memories as a 3D "Nebula." The intensity of a memory node $M$ is calculated based on emotional resonance $E$ and recall frequency $R$:$$M = \sum (E_{i} \cdot R_{i})$$ Families could share and watch the life-journey videos together.
Predictive Health: Using $dt$ time-series analysis on cognitive trends to alert doctors to early signs of decline before a crisis occurs.

Before time forgets, let the light remain.