Logging claude code file access with MCP

I recently read this post about a CLI agent that uploaded local files to the cloud, without anyone asking it to. That’s one of the biggest risks with coding agents without traceabilit...

Automatic Skill Generation for Web Agents

Hallucination detection without LLM

Synthetic Textile Data with Blender and Generative AI

This blog shares the latest version of our synthetic-data pipeline for generating better training data with Blender. We render a synthetic view of the textile on a table from differen...

Prompt Evolution

This blog aims to develop an abstract framework to store prompts of agentic functions in a graph database (considered Neo4j for this example). These prompts will be versioned over cha...

Security layer for mcp server

This started as an internal project where we needed a structured, reusable security layer for an MCP server we built to expose database queries and data to different user roles such a...

Improve small language model monitoring in production

I read the paper Pioneer Agent and found several interesting ideas that are highly relevant for building agentic workflows in production systems. We have used some learning in our pro...

Can a Gemini LLM beat StepLR for ResNet-18 on CIFAR-10?

I wanted to see if a Gemini LLM, called once per epoch, could pick a better learning rate than a fixed StepLR schedule for ResNet-18 on CIFAR-10.

Zero trust data pipeline framework

We have all experienced the pain of data loss or data incorrectness in our batch data pipelines. To address this, I designed a framework for ensuring data correctness at every stage o...