I'm a Founding Engineer & AI Lead at Storefox.ai, building production multimodal LLM systems for retail intelligence.

Alongside engineering, I work on technical AI safety research — particularly AI control, agent security, model evaluations, chain-of-thought faithfulness, and model misalignment.

• 🛡️ Janus — Open-source security framework for LLM agents, enforcing least-privilege policies at the tool-call boundary. Presented at Black Hat Arsenal, SecTor 2026.

• 🧠 Multimodal CoT Faithfulness — Whitebox Research study investigating how faithfully models explain their reasoning across text and image inputs.

• 🔍 LLM Self-Recognition — Replication experiments exploring whether language models recognize and prefer their own generations.

• ⚙️ Transformer from Scratch — A modular PyTorch implementation of the original Transformer architecture.

• AI control and monitoring autonomous agents
• Reasoning vs. agent-trace monitorability
• Shutdown resistance and emergent misalignment

Tools I reach for: Python · PyTorch · LLMs · AWS · Docker · Git

Portfolio · LinkedIn · Technical Writing · Substack

>>> ishita.generate("Can we work together?")
'Yes! LinkedIn or email are the fastest ways to reach me.'

Usually reading, listening to music, or wondering whether an LLM's explanation explains anything at all.