Recall | 🏆 2nd Best Overall @ HackUSF 2026

Links 🔗: Devpost | Demo Video

Team: Stevin George, Sebastian Noel, Nicole Bustos, Chris Ho

Summary

Recall is a wearable AI memory assistant integrated into glasses, designed to support individuals with cognitive impairment or memory loss. It restores autonomy by acting as a passive, "always-on" observer that remembers daily details on the user’s behalf.

• Passive Capture: Continuously records visual and audio data.

• Natural Language Queries: Users can ask questions like "Where did I leave my keys?" or "Did I take my medicine today?" and receive timestamped audio answers.

• Semantic Search: Uses AI to understand the meaning of scenes, allowing for intuitive memory retrieval.

Hardware

Built using two ESP32 microcontrollers mounted directly onto glasses frames:

• ESP32-S3: Handles video via an OV2640 camera module, streaming MJPEG frames over WiFi.

• ESP32-WROOM-32E: Captures 16kHz mono audio using an INMP441 I2S MEMS microphone.

• Connectivity: Custom multi-endpoint HTTP and TCP socket streaming to a local processing pipeline.

Software

• Vision Pipeline: Processes video batches using YOLOv8 for object detection and Gemini 2.5 Flash for rich, spatial scene descriptions.

• Memory Engine: Stores data in SQLite and uses ChromaDB vector embeddings to enable "meaning-based" (semantic) search.

• Audio & Voice: Transcribes environmental audio via ElevenLabs Scribe and delivers verbal responses using ElevenLabs TTS.

• Dashboard: A Next.js web interface featuring a live memory timeline, mood analytics, and hardware health monitoring.

Tech Stack

• AI/LLMs: Google Gemini 2.5 Flash, ElevenLabs (Scribe v2 & TTS).

• Computer Vision: Ultralytics YOLOv8, OpenCV.

• Databases: ChromaDB (Vector DB), SQLite.

• Frontend: Next.js 16, TypeScript, Tailwind CSS, Recharts.

• Backend: Python, Flask (Multi-threaded architecture).

• Firmware: C++ (ESP32).

---

Dashboard

Database