article4 min read
Vetting a Fresh Open Weights Coding Model Before It Touches Your Codebase
What to check in a new open weights coding model's license, GPU sizing, agent sandbox, and eval results before it reaches production code.
← All posts
writing · 23 posts
Articles
Guides, deep dives and opinion pieces on LLM systems, Postgres, queues and deploy pipelines.
- article7 min read
Reading GPT-6.5 Leaks Before a Codename Reversal Costs You a Migration
A tiered method for weighing GPT-6.5 rumors, a dated look at one reversed leak, and an alert setup that waits for API evidence.
- article7 min read
Before You Upgrade Your Model Build an Eval Harness From Your Own Prompts
How to read recent model release news, track regressions against a baseline, and run a small offline eval before migrating.
- article6 min read
Why Cheaper Models Can Win on Cost per Completed Task
How to compare LLMs by cost per completed task, and route work between Haiku and Sol with validated escalation.
- article8 min read
What the GPT-7 Bel Rumor Changes for Your Model Migration Plan
How to separate verified GPT-7 and Bel claims from rumor, and prepare model migrations with evals, pinning, and staged rollouts.
- article9 min read
Why We Picked a Polling Outbox Over Debezium for 4,000 Writes per Second
How we weighed a polling transactional outbox against Debezium CDC for order events, with the tradeoffs and production numbers.
- article4 min read
Running Claude Haiku 5.5 for High-Volume Ticket Triage
What Claude Haiku 5.5 costs, what its reported benchmarks show, and how to call it from Python for cheap, high-volume classification.
- article3 min read
Building State Machines That Survive Missing Events
Handle out-of-order and lost events in distributed systems without strict ordering. Practical patterns for gap detection, recovery, and avoiding zombie states.
- article3 min read
Cost Attribution in Multi-Tenant GenAI Systems
Practical instrumentation strategies for measuring and allocating LLM costs in production across tenants, A/B tests, and feature flags without sacrificing latency.
- article4 min read
Designing Fallback Chains When LLM Requests Fail Mid-Stream
Practical patterns for handling LLM failures at scale: cached responses, tiered model fallbacks, request reshaping, and honest trade-offs when your primary strategy breaks.