WRITING & STRATEGY

Blogs & Playbooks

Practitioner-level writing on enterprise AI strategy, architecture, and deployment — from 8+ years building production LLM systems.

Deep Dive Playbooks

Aug 20, 2026·13 min read·FEATURED

Making Cancer Vaccines the Default for Dogs

Moderna proved the science. YC funded a competitor. Neither solved the part that decides who wins.

A backcast from 2031 on what it would take to make personalized mRNA cancer vaccines the standard of care for dogs in North America. Where the money actually goes in a personalized dose (manufacturing is 70–90% of it, sequencing is a rounding error), how one Australian did it for $3,000 by never entering the commercial manufacturing market at all, the seven things that had to go right, the three ways it dies, and a strictly ordered year one that puts the co-founder and the CDMO ahead of more pipeline code.

Veterinary OncologyBiotechStrategyCancer Research

Read

Aug 5, 2026·14 min read·FEATURED

So What The Hell Is A Harness?

DeepSeek V4 Flash is 90x cheaper than Opus 4.8 and good enough. I still haven't switched. Here's the layer nobody benchmarks.

Eight out of ten devs I know haven't moved to the cheap open models, and the reason isn't the model. A full dissection of the agent harness: the seven layers wrapped around the weights — context assembly, tool design, loop control, compaction, permissions, verification, recovery — with the failure mode each one produces. Plus the L0–L5 hierarchy of modern coding, why the L2→L3 jump is a harness jump and not a model jump, and the circularity buried in DeepSeek's own benchmark footnotes: several headline scores were run on an internal harness they never released.

Agent HarnessAgentic AIClaude CodeDeepSeekDeveloper Tools

Read

Aug 1, 2026·16 min read·FEATURED

Hey DeepSeek, I Fixed Your Branding

DeepSeek-V4-Flash-0731 is peak unreadable. Here's the whole naming system and mascot line, rebuilt.

DeepSeek ships some of the best open-weights models on earth and names them like industrial fasteners. A full unsolicited rebrand: a Deep 1–4 generation ladder with role suffixes, one whale mascot that visually evolves across generations, and every model DeepSeek has actually shipped since 2023 remapped onto it — including the Deep 3.5 tier the history forces you to add, and the argument for why reasoning should be a toggle instead of a SKU.

BrandingProduct NamingDeepSeekDesignOpen Source

Read

Jul 26, 2026·16 min read·FEATURED

Cut the AI Chatbot Tax Before It Cuts You

Every engineering leader shipped the assistant. Now the CFO wants the cloud bill explained — here's the 90% fix.

Vectorizing your company's docs, Slack, and codebase can 16x your data footprint overnight. A worked case study on where the memory actually goes, the two levers most teams pull first, the object-storage-first architecture that cut vector search costs by up to 95% for some teams, and the two S3 features from 2020 and 2023 that made it possible.

Cost OptimizationVector SearchTech ArchitectureAWSAI Infrastructure

Read

Jul 25, 2026·22 min read·FEATURED

The Technology Stack Behind Production RAG + MCP

From S3 and DynamoDB to a Citation-Grounded MCP Server — with an Azure/Databricks Mapping

A companion to the RAG system design playbook — this one covers the infrastructure those decisions run on. Concrete AWS services, data schemas (DynamoDB items, OpenSearch mappings, MCP tool schemas), governance, observability, and a clean mapping to Azure/Databricks.

Tech ArchitectureRAGMCPAWSData Engineering

Read

Jun 19, 2026·40 min read·FEATURED

The AI Compute Supply Chain: Mapping Every Bottleneck From Sand to Inference

A field guide to AI infrastructure — what training and inference actually demand, the physical chain behind them, and where it's bottlenecked in 2026

First, what actually goes into building and running an AI model. Then the entire physical supply chain behind it — copper, lithography, CoWoS packaging, HBM memory, optical interconnect, grid power, cooling, and neoclouds — with who benefits at each stage, where it's bottlenecked in 2026, and a priced-in / room-to-grow / contested verdict on the public companies exposed. Every technical term defined in plain language first.

AI InfrastructureSemiconductorsSupply ChainMemoryEnergyInvesting

Read

May 22, 2026·25 min read·FEATURED

Designing a GenAI News Discovery System: Every Decision, Every Tradeoff

A System Design Case Study for CS Grads — RAG, Editorial AI, and Production Readiness

Walk through every architectural decision in building a GenAI content discovery system for a major news media corporation — from RAG vs fine-tuning to chunking strategy, hybrid retrieval, editorial guardrails, and evaluation. Includes a live interactive demo.

System DesignRAGVector SearchEditorial AILLM

Read

Apr 21, 2026·15 min read·FEATURED

Project Rosie: An Open Source AI Pipeline for Personalized Cancer Vaccine Design

How Gemma 4's agentic orchestration, multimodal vision, and 256K context make it irreplaceable — not bolted on

Deep dive on Project Rosie: the full pipeline architecture, five distinct Gemma 4 integration roles (agentic orchestrator, multimodal visualisation interpreter, clinical report generator), and why veterinary oncology is the right launch market.

Health & SciencesGemma 4Veterinary OncologyOpen Source

Read

Apr 09, 2026·17 min read·FEATURED

A man used ChatGPT and AlphaFold to design a cancer vaccine for his dying dog. It worked. Here is how to replicate it.

How Rosie proved the pipeline works

A deep dive into the Rosie case, the end-to-end neoantigen pipeline, and why veterinary oncology is the fastest path to build open personalized mRNA vaccine infrastructure.

BiotechCancer ResearchOpen SourceAI

Read

Mar 17, 2026·22 min read·FEATURED

Enterprise Agentic AI: From Chatbot to Autonomous Agent

A Practitioner's Guide to Production-Grade Agentic Workflows

What changes when you move from a chatbot to an autonomous agent. Deep-dives into MCP, LangGraph orchestration, memory architecture, human-in-the-loop design, and the Padala Bank Credit Analyst Agent as a reference implementation.

Agentic AIMCPLangGraphEnterpriseOrchestration

Read

Mar 10, 2026·18 min read·FEATURED

Enterprise GenAI Strategy: From Data to Production

A Complete RAG-Based Knowledge System Rollout Playbook

Why most enterprise AI projects fail — and the phased, opinionated playbook for getting RAG-based knowledge systems right. Covers data pipelines, RBAC, evaluation, change management, and governance.

RAGEnterprise AIData ArchitectureRBACStrategy

Read

External Writing

The AI Compute Supply Chain: Mapping Every Bottleneck From Sand to Inference

Personal

Designing a GenAI News Discovery System: Every Decision, Every Tradeoff

Personal

Enterprise Agentic AI: From Chatbot to Autonomous Agent

Personal

Enterprise GenAI Strategy: From Data to Production

Personal

Product Management 101

Medium

Project Management Zero to Hero

Medium

ZeroSig semi-finals at Great Canadian Web3 Pitch Showdown