Feb – Jun 2026 · TwelveLabs · Shipped to beta
Electra
An internal AI agent built on top of the company’s IT ticketing system. I owned most of the implementation alongside my manager, and shipped it to beta before my term ended.
Internal tool — no public repo.
- Memory
- Every conversation, interaction, and thought links into a graph-based knowledge network — prototyped in Obsidian, then migrated to GBrain, our internal graph memory. Context persists across sessions and the reasoning stays traceable.
- Defense
- Hardened against prompt injection, jailbreaks, and token-bomb exploits: rate limiting, input sanitization and validation, RBAC-style permissions, and anomaly detection with logging.
- Next
- Designed the foundational architecture for a Hermes-style self-learning structure, so agent behavior improves on its own over time.