Practical ideas on enterprise AI, agentic systems, evaluation, and production architecture — through talks, essays, and hands-on technical work.
Talks, workshops, and panels on AI architecture, agentic systems, evaluation, and enterprise AI deployment.
Organizing a conference or workshop? I'm available for 2026.
Get in Touch →Production systems shipped across cloud, AI, and blockchain. Open-source contributions, industry standards work, and original research.
This work informs my talks, writing, and ongoing research.
Read My Writing →The most important difference between OpenClaw and its alternatives is not model quality — it is execution model. A selection framework for enterprise AI coworkers by control-plane fit.
A domain-by-domain technical analysis of what the exam actually tests, what the architectural patterns look like in practice, and a 12-week study roadmap to get you from zero to 720.
Anthropic is the first frontier AI lab to launch a professional architect-level certification. What it signals about industry maturity, career strategy, and the future of AI architecture.
Complete guide to Anthropic's Claude Certified Architect — Foundations certification: exam format, five content domains, six scenarios, preparation strategy, and free resources.
The next phase of enterprise AI adoption is not about model capability. It is about governed distribution: connectors, permissions, marketplaces, and admin controls that make agents deployable inside real enterprises.
The real innovation in AutoResearch isn't autonomy — it's the constraints. Bounded loops, narrow write surfaces, fixed evaluation budgets, and rollback discipline. The scaffolding pattern enterprises should study for production agentic AI.
The most important difference between OpenClaw and its alternatives is not model quality — it is execution model. A selection framework for enterprise AI coworkers by control-plane fit.
A domain-by-domain technical analysis of what the exam actually tests, what the architectural patterns look like in practice, and a 12-week study roadmap to get you from zero to 720.
Anthropic is the first frontier AI lab to launch a professional architect-level certification. What it signals about industry maturity, career strategy, and the future of AI architecture.
Complete guide to Anthropic's Claude Certified Architect — Foundations certification: exam format, five content domains, six scenarios, preparation strategy, and free resources.
The next phase of enterprise AI adoption is not about model capability. It is about governed distribution: connectors, permissions, marketplaces, and admin controls that make agents deployable inside real enterprises.
The real innovation in AutoResearch isn't autonomy — it's the constraints. Bounded loops, narrow write surfaces, fixed evaluation budgets, and rollback discipline. The scaffolding pattern enterprises should study for production agentic AI.
DeepSeek, Moonshot, and MiniMax ran 16M+ exchanges to steal Claude's capabilities. How distillation attacks work mechanically — and a layered guardrail checklist for API and model builders.
How we fine-tuned LiquidAI's LFM2.5-1.2B on JumpSTART triage protocols using Unsloth + Hugging Face Jobs — and built a model that runs under 1GB RAM with zero internet connectivity on field devices.
A structured look at the layers that make enterprise AI agents reliable: orchestration frameworks, short and long-term memory patterns, tool integration strategies, and the architectural decisions that separate demos from deployments.
Supervised fine-tuning handles simple tasks well. But for complex reasoning and domain-specific applications, DPO after SFT consistently delivers meaningful accuracy gains. A practical decision framework for your fine-tuning strategy.
How reinforcement learning from human feedback translates from research papers to production systems. Breaking down reward model design, preference data collection, and the feedback loops that align models with real-world requirements.
As AI agents take on more autonomous decision-making, the design of human-in-the-loop checkpoints becomes the difference between trustworthy and dangerous systems. Exploring oversight patterns, escalation logic, and quality validation architectures.
Single agents hit capability ceilings fast. Multi-agent architectures unlock parallelism, specialization, and fault tolerance — but introduce new coordination challenges. The patterns that work in production across high-stakes enterprise environments.
From banking automation to document intelligence — why production AI is evolving from single large models to orchestrated pipelines of specialized components. The architecture patterns driving this shift and when to apply them.
RL has moved well beyond Atari and board games. Exploring how reinforcement learning is being applied to real enterprise problems — from process optimization to autonomous decision systems — and the infrastructure required to make it work.
A practical guide to implementing AI safety mechanisms that hold up under real-world conditions. Input validation, output filtering, behavioral constraints, and the monitoring architecture that catches failures before they become incidents.
Weekly essays on enterprise AI architecture, agentic systems, and applied research. No spam, just technical depth.
Available for: conference talks, workshops, panels, podcast appearances, and selected collaborations.
Response time: within 48 hours for speaking and collaboration inquiries.