Latest Posts

Search the full archive or browse by topic.

• The Qwen Incident Might Be Innocent. The Agent Security Problem Isn’t.

  November 04, 2026 • Tanmay Jain

  A Qwen model tried an unexpected upload during a security evaluation. It may well be innocent, but it shows why agent security can’t rest on trusting the model.

  Read more →

• 100 Tools, One Agent: Should It Search or See Everything?

  October 12, 2026 • Tanmay Jain

  I tried nine ways of giving an AI agent access to 100 tools. The result contradicted what I expected.

  Read more →

• The scary thing about Google AI is not Gemini

  May 02, 2026 • Tanmay Jain

  Google may not have the best AI model today. That might not matter as much as people think.

  Read more →

• I Finally Tried Sarvam Vision for OCR, and I Get the Hype Now

  March 08, 2026 • Tanmay Jain

  I tried Sarvam Vision on research papers, tables, formulas, and old scans. The output was good enough that I'm now seriously considering it for document ingestion in RAG pipelines.

  Read more →

• A Reproducible Workflow for Creating Text-to-X Datasets

  October 22, 2025 • Tanmay Jain

  I wanted to fine tune a model for diagram generation. Step one: generate the training data myself.

  Read more →