John.I.

AI | Energy| Creator@gridmix

[John.I. @johnigwebuike]

Evals, the new PID

Models are cheap

Prompts are easy

Demos are everywhere

The hard part is knowing whether your system is actually improving

In control theory, feedback creates stability.

In AI, evals create progress

Without evals, you're tuning it by vibes

#AIEngineering

[John.I. @johnigwebuike]

versioning, Evals and observability are is all you need.

https://t.co/TcAZ6dd8Lb

[John.I. @johnigwebuike]

@Vinayak_M11 Sure @Vinayak_M11

[John.I. @johnigwebuike]

So teams hack together:

spreadsheets
logs
manual reviews
intuition

[John.I. @johnigwebuike]

DM me

[John.I. @johnigwebuike]

Especially if you’ve thought:

“LLM workflows feel unstructured and hard to improve”

[John.I. @johnigwebuike]

Which means:

We’re trying to optimise systems
we can’t properly measure or observe

[John.I. @johnigwebuike]

LLMs are non-deterministic systems.

So:

outputs vary
regressions are subtle
behaviour drifts over time

[John.I. @johnigwebuike]

The real problem isn’t writing prompts.

It’s that we can’t reliably:

improve them
evaluate them
or understand how they behave

[John.I. @johnigwebuike]

Prompt engineering isn’t engineering.

It’s still trial-and-error with better tooling.

[John.I. @johnigwebuike]

Every team using LLMs ends up doing this:

tweak prompts
rerun manually
compare outputs
guess what improved

It’s messy.

[John.I. @johnigwebuike]

Evals are all you need

https://t.co/BxghdSPEdP

[John.I. @johnigwebuike]

The power of system prompts.

[enprompta @enprompta]

Discover the Hidden Power of System Prompts — the invisible instructions that shape how AI actually behaves behind the scenes. 🤖
Must-read for prompt engineers & AI builders 👇
https://t.co/cr4c9xvxU2

[enprompta @enprompta]

Discover the Hidden Power of System Prompts — the invisible instructions that shape how AI actually behaves behind the scenes. 🤖
Must-read for prompt engineers & AI builders 👇
https://t.co/cr4c9xvxU2

[John.I. @johnigwebuike]

Message anyone.
No internet required.
End-to-end encrypted.

NearChat is coming soon.

https://t.co/pgD6zhOGmm

#MeshNetworking #OffGrid #Decentralized #Web3 #Privacy #P2P #BluetoothMesh #EmergencyPrep #DisasterRecovery #GridDown #Prepper #SurvivalTech #Resilience #Hiking #Backp

[John.I. @johnigwebuike]

Building something for when the internet fails you.

NearChat: Message anyone. No internet required.

Uses Bluetooth mesh to keep you connected during:
- Natural disasters
- Remote adventures
- Network outages

Join the waitlist → https://t.co/ouYmyeIVcC