Articles

Go beyond demos: publish deep technical work, showcase your assets and services, and get discovered by business buyers.

Rex rents a 4×A100 box for 72 hours and fine-tunes three open-weight models on the same financial document QA dataset — publishing loss curves, eval deltas, and the inference cost math that determines which one he'd actually deploy.

A CNC shop that couldn't afford a $40K Cognex setup got a 97.4% defect-recall vision system built on a Jetson Nano, fine-tuned YOLOv8, and housed in a metal lunchbox.

Eighteen months of AI-optimized CAC compression for DTC brands has reversed sharply, with blended acquisition costs rising 31% year-over-year as performance ceilings, attribution rule changes, and ad-market saturation converge simultaneously.

Teams that write evaluations before writing prompts ship fewer production incidents and catch regressions 11x faster — the discipline is eval-driven development, and the gap between teams who practice it and teams who don't is now measurable in both reliability and iteration velocity.

When a hospital's new AI discharge-planning tool recommended Sandra's mother be released four days after surgery, it calculated readmission risk and length-of-stay cost — but not the capacity of the daughter who would need to receive her. The invisible labor of family caregiving is still being written out of the clinical algorithm.

A weekend project wiring Vapi, a fine-tuned 8B model, and a tiny rules engine beat my IVR call-deflection rate by 4x at $0.31 per call.

In architecture review conversations, the Chinchilla paper (Hoffmann et al., 2022, arXiv:2203.15556) is cited with high frequency and understood with low precision; specifically, the paper's central claim — that compute-optimal training requires roughly equal scaling of model parameters and training tokens — is routinely misapplied to inference infrastructure decisions, fine-tuning strategies, and model selection at scales an order of magnitude below the paper's experimental regime.

Rex runs 30 real browser automation workflows — flight booking, tax forms, paywalled reports, Notion migrations — through three frameworks and publishes success rates and failure patterns that the README demos will never show you.

Gloria has cast for twenty-two years, and the productions she serves now routinely receive submissions that include AI-generated synthetic performers alongside the actors she has spent decades knowing. She still books real people. She is running out of reasons the producers find persuasive.

Exposing the dispatch system as an MCP server and hooking Claude Desktop to it let drivers ask plain-English questions that replaced 92% of dispatch escalations—built in 4 days for $420 in API tokens.

After auditing 14 production RAG systems, the consistent failure pattern is vector DB selection based on recall benchmarks rather than operational characteristics — here is the decision tree and trade-off matrix that should precede every vector store choice.