Writing

Blog.

Engineering writing from the Veso team. Architecture, production playbooks, and field notes on shipping AI into legal, healthcare, and enterprise workflows.

Archive

All articles

2026
  1. A harness for a model that only chooses Jev picks from a list and returns calibrated probabilities. It cannot write, plan, or read pixels. We built a loop around those limits, measured it driving a Mac, and found the model was the easy part.
  2. AI Accelerator Options in 2026: NVIDIA, AMD, Google, AWS, Cerebras, Tenstorrent Twelve accelerators on memory per dollar, bandwidth per compute, MLPerf tokens per dollar, domain size, and software status.
  3. Kimi-Linear-48B on Tenstorrent Wormhole Veso AI open-sources the fused KDA, MoE, and MLA kernels that took Kimi-Linear-48B from not starting at all to a 16K-context server on four n300 cards.
  4. The Classifier That Watches Your Coding Agent Spawn a subagent in Claude Code and a second model reads the call before it runs, scores it 0 to 100, and blocks anything above 50. The full ruleset, captured off the wire and reproduced here in one piece.
  5. Knowledge as Dynamics: The Paper Veso R&D releases Knowledge as Dynamics: distilling, composing, and inheriting the latent vector fields of neural networks. Five laws, one teacher-only instrument, every number reproducible on a laptop.
  6. The LLM Timeline: What Lived and What Died Every pivotal LLM release from GPT-2 to Kimi K3, the six eras they define, and the techniques each era killed.
  7. Kimi K3: A Forensic Analysis A claim-by-claim audit of Kimi K3: architecture, benchmarks, pricing, and hardware.
  8. What We Learned Validating JEPA on a Laptop We reproduced the core claims of two 2025-26 JEPA papers (CrossJEPA and LeJEPA) at toy scale on an M5 MacBook Air. The mechanisms hold. Here is what that means for enterprises deciding whether self-supervised representation learning is real or hype.
  9. The Agentic Harness: Why the Orchestration Layer Is the Product Models are commodities. The harness, the control plane that governs what an LLM sees, calls, and outputs, separates demos from production AI systems. A technical breakdown of the architecture pattern defining 2026.
2024
  1. Reciprocal Rank Fusion (RRF) Explained Unlock the power of Reciprocal Rank Fusion in Retrieval-Augmented Generation. Learn how RRF combines results from multiple retrieval methods in RAG.