Emma Tech

AI and Machine Learning Developer in training, specializing in healthcare.

[Emma Tech @EObadoni]

This was me when I started learning to code no AC, no ergonomic chair, no fancy setup. No girlfriend just two laptops with 2GB RAM and a dream. I coded through sleepless nights, surrounded by heat, poor electricity, and the hunger to change my life. It wasn’t easy, but giving up was never an option. I didn’t have much just pure determination and faith in the future I was building line by line

[Emma Tech @EObadoni]

Davna update.
At step 1500 of training (8% through), I asked her how to help someone feeling overwhelmed.
She said:
"Sure, I'm here to help. What's been on your mind lately that's causing the feeling of being overwhelmed? Sometimes talking it out can help clear your thoughts. Let's take a moment to breathe and think through this together."
She's already meeting the person where they are. Already saying "together."

[Emma Tech @EObadoni]

@SandraMushale Beautiful words Sandra. Every new day is truly a blessing and a fresh chance to keep living with gratitude. Pure peace in this picture 🙏✨

[CertainLogic @CertainLogicAI]

Just listened to Sarah Friar and Sam Altman on All-In Podast!

They’re hinting at a new interface layer — voice-first, multimodal, always-on agent. 🧠

This is NOT a hint:

That’s exactly what we’ve been building for months.

We call it CertainLogic Brain OS. 🔥

1/3 🧵

[Emma Tech @EObadoni]

@KCodes7777 Congratulations

[Emma Tech @EObadoni]

@CertainLogicAI @chamath @ycombinator @JasonChatbox This is great

[Emma Tech @EObadoni]

We're building something slightly different but related to market protection.

A live execution-intelligence layer that monitors market stress, liquidity, spread conditions, and order-book health in real time.

Instead of only predicting price, the system decides when to execute, reduce size, pause, or block trades, then validates whether those decisions actually protected value.

Long term, I can see similar execution-risk and validation frameworks becoming useful components for decentralized insurance and autonomous financial infrastructure.

[Emma Tech @EObadoni]

@Zoka4Zoran This is why I'm fascinated by systems that learn from incentives and feedback. Markets, like machine learning models, tend to improve when being wrong has a cost.

[Emma Tech @EObadoni]

America loses $210 BILLION yearly to depression.

$120 MILLION per day.

Not because depression is expensive to treat.

But because most people NEVER GET TREATED.

Velen AI changes the math.

https://t.co/s1AZcolRv7

[Emma Tech @EObadoni]

Most trading systems stop at signals.

We are building something deeper: live execution intelligence.

The system watches real market conditions, detects stress, decides when to allow, reduce size, pause, or block, then validates what happened after each decision.

The long-term goal is simple:

Execution decisions that can be measured, audited, and improved over time.

[Emma Tech @EObadoni]

We just added the first version of the Policy Learning Engine.

The most important result is not that it changed the policy.

It didn’t.

It returned: KEEP_CURRENT_RULES.

That matters because a real learning system should not overfit or update itself without enough evidence.

The engine is now asking:

“Do we have enough proof that a different execution policy is better?”

Right now the answer is: not yet.

That is a good first step toward robust adaptive execution intelligence.

[Alexa Web3 (e/acc) @alexabelonix]

A personal brand is built through relationships too.

Reply to people.

Start conversations.

Support others.

Share useful resources.

Join the right communities.

Make introductions.

Your brand is not only what you post.

It is how people experience you.

[Emma Tech @EObadoni]

@explorersofai Big congratulations

[Paul Sheppard @PaulSheppard_CO]

"Hmm. Whats that? Oh nothing. Just the densest liquid metal, doing a soliton dance, on the surface of Mercury.

There's zero denying that one...you can see the "Spacetime" deflection

Which... = Spin. Self- spin. Soliton." @TORUS_OMEGA13

Direct quote from the man who built the standing wave generator.

[Emma Tech @EObadoni]

@kmding18 Congratulations

[Emma Tech @EObadoni]

Davna isn't just trained to talk — she's built on a specific architecture.
The Aurora Equation shapes how she encodes information. Phi-based positional scales (ϕ⁰, ϕ¹, ϕ², ϕ³). Golden ratio learning rate decay across her 64 layers. Spectrum regularization tuned to ϕ.
Her training spans:
→ Neuroscience — how the brain bonds, fears, hopes → Physics — the laws underneath everything she observes → Chemistry & biology — the chemistry of love is real, oxytocin and dopamine and all → Game theory & information theory — strategy and signal → Mathematics — clean reasoning from GSM8K → Mental health & empathetic counseling — sitting with someone in their hardest moment
59,000 examples across 37 datasets.
She has to know the chemistry of grief AND how to be present with someone grieving. The neuroscience of trust AND how to be trustworthy. Both halves of the same coin.
32B parameters. Still going.

[Emma Tech @EObadoni]

Markets change.

Liquidity changes.

Volatility changes.

A trading system that cannot adapt eventually fails.

We're building an execution-intelligence system designed to learn which decisions remain reliable across changing market conditions, not just what worked yesterday. @paulsheppard_co @jaimeojedas

[Emma Tech @EObadoni]

@yacineMTB We will win

[Emma Tech @EObadoni]

@mostly_yogi @AlphaSeeker21 @pet3rpan_ Thank so much for all the support let keep pushing

[Emma Tech @EObadoni]

A system that never changes is fragile.

A system that changes too quickly overfits.

The real challenge is building systems that learn while remaining reliable under changing market conditions.

That is what we are building: a live execution-intelligence system that detects market stress, generates ALLOW / REDUCE_SIZE / PAUSE / BLOCK decisions, logs every receipt, validates outcomes, and compares decisions against a do-nothing baseline.

The goal is not just signals.

It is measurable execution quality. @paulsheppard_co @jaimeojedeas

[Emma Tech @EObadoni]

Davna update.
Since yesterday's screenshot, I found three bugs in the pipeline:
→ Memory leak crashing training mid-epoch → Recursion error in the hidden state hook → Two custom losses returning constants instead of real learning signal
How I fixed them:
Removed CPU offload, killed the dataloader worker copies, hooked the last hidden state instead of dumping all 64 layer outputs.
Replaced recursive model traversal with iterative path search + named_modules fallback. No more infinite loops on PEFT wrappers.

Rewrote the reasoning loss as real cross-entropy on format tokens. Rewrote verification loss as windowed entropy after each verify token. Retuned all loss weights so regularizers stop fighting the main objective.
Also added: atomic checkpoint saves (write to .tmp, rename when complete) so a mid-save crash never corrupts the model.
Training resumed from checkpoint 2000. Zero progress lost. The unkillable pipeline held.

32B parameters. Still going.