One thing I’m learning with node workflows like Flora and Weavy is that each step needs to be specialized. Then you layer in the specifics.
A lot of what I’m building right now is for branded imagery. With image models, you get plenty of unwanted artifacts, strange details, and gooey limbs, so it can take a few passes to get the final image where you want it.
I’ve always admired art directors for their eye for detail. These tools force you to put that hat on.
The thing I wish these tools had is a better way to draw, mark up, and revise as you move through the workflow.
In any case, I’m really proud of the images that are coming to life.
1
2
120
As a creative strategist - using ai in my workflow greatly cuts down on the back and forth.
By integrating what I call an AI "rough-cut" or "pre-visualization" of the concept I am briefing for - I am able to provide a LOT more clarity to either the creator or AI engineer I am handing off to in the next step.
To be able to do this - I need to understand how to generate accurate UGC quality videos myself - so that the AI roughcut is close enough to what i want the intended final deliverable to be.
The role of creative strategist is increasingly becoming AI generalist 👩🏼💻 🤖
0
285
The problem. Teams ship LLM features for contracts without knowing how accurate they are or when a prompt change breaks them.
What I built. A multi-provider eval harness on CUAD contracts (150-example frozen set, 6 clause categories) with bootstrap confidence intervals, an LLM judge validated against human labels, calibration metrics, and a GitHub Actions gate that comments on every PR.
Results. Clause presence F1 0.88-0.90 (95% CI), judge agreement κ = 0.754, ECE 0.053-0.100, 0% parse errors. Found a shared failure mode: all three frontier models truncate the legally operative condition at span boundaries.
Stack. Python, FastAPI, Next.js, Cloud SQL, GitHub Actions, Bedrock, Gemini, Cloud Run.
1
64