https://www.linkedin.com/in/jesse-sng-05ab7944/
Imagen is cheaper for image generation than nano banana but it does not support text. if text is required, gpt image 2 would be still affordable. you can render and super impose text over image to save cost. https://deepmind.google/models/imagen/
When you do parallel workflow, always plan on orchestration and piecing things together
Error management if there is a failure, plan the workflow to continue first then go back to retry. Do not discard the the previous effort.
image cursive improvement —> try comftyUI https://comfy.org/
video wise, to consider remotion https://www.remotion.dev/
observability
grafana for light work https://grafana.com/
prometheus for heavier stuff https://prometheus.io/
https://github.com/gauravmm/2026-06-Evals
https://www.linkedin.com/in/gauravmanek/
evals-for-ai-agents-slides.pdf
Eval vs Guardrails Both can be overlapping Guardrail can be done by a deterministic code
Another way is to limit the blast radius by infrastructure (separate mac mini/ cloud/etc) or via action (draft email over sending email, mark email over delete emails)
SGInnovate
PIC: An Ting linkedin.com/in/antinglim/?skipRedirect=true
⏳ Time and Mic Keeper - Jasmine https://www.linkedin.com/in/kaimingng/
📸 Memories Keeper - Esther https://www.linkedin.com/in/esthertanshuzhen/
🎙 Host - Han
🧠 Knowledge Keeper - Han