Give the fast path its own lane.
I separated live captioning from the heavier extraction and graph-write path. The display could stay responsive while structured memory was built in the background.
1stOpen Track · JacHacks SF 2026
~5 hrsFrom idea to working demo
$1,500Team prize
Start with the
actual problem.
At a busy event, conversations move faster than memory. Someone is building a company. Someone else is looking for an introduction. The useful detail often disappears before you can do anything with it. We explored whether smart glasses could preserve that context while keeping you in the conversation.
The work behind
the outcome.
Model relationships, not a pile of notes.
People, companies, topics, and intent became graph nodes and relationships in Jac. Walker agents could traverse those connections to retrieve useful context.
Keep the wearable client thin.
The Android/Kotlin client focused on capture and display. The memory logic lived behind it, reducing the amount of integration work on unfamiliar hardware within a five-hour window.
How the pieces connect.
Speech capture
FAST PATHLive captionsWearable display
BACKGROUND PATHLLM extractionRelationship graphRecall card
Explanatory system view · reconstructed from the project
RECALL / MEMORY ENGINE
01 — BE IN THE CONVERSATION
“I’m Jacob, a startup founder.The conversation happens. Context starts taking shape.
I’m building a new venture and looking for funding.
Could you make an investor introduction?”
SAMPLE DATA · RECONSTRUCTED INTERFACE
Hands on.
People with me.
I helped shape the product, built the graph-based Jac backend and extraction flow, and worked on the Android/Kotlin integration. The working wearable demonstration and hackathon win were team efforts.
Jac / Jac CloudGraph traversalLLM extractionAndroid / KotlinMeta Ray-Ban Display
What came out of it.
We built and demonstrated Recall on Meta Ray-Ban Display smart glasses in approximately five hours. The team won first place in the Open Track at JacHacks SF 2026 and a $1,500 prize.
CONTEXT & LIMITS
Recall was a hackathon prototype, not a production memory service. The interactive example here reconstructs the idea using illustrative people and data; it is not original product footage.
Qualcomm
See the demand. Make the better call.