Alejandro Espinosa

Senior electronics engineer building practical AI systems for real-world operations. I turn domain expertise into useful tools.

[Alejandro Espinosa @alejandro_spi]

I’m Alejandro, a senior electronics engineer building AI systems for the real world.

I’m not interested in AI as content or demo-only workflows.
What interests me is this:
taking real operational knowledge from people who live the work every day and turning it into useful systems.

That’s the kind of work I’m doing now:

• practical AI tools for real business workflows
• agent-based systems shaped by domain expertise
• software for engineering, operations, and automation
• projects like NormaLang and CoordinIA

I’ll use this account to share:

• what I’m building
• what actually works with AI agents
• mistakes, tradeoffs, and edge cases
• lessons from turning real-world knowledge into practical tools

If you care about practical AI, real workflows, and building things that survive contact with reality, you’ll probably
like this account.

[Alejandro Espinosa @alejandro_spi]

@catrina_nortena @Claudiashein @MaruCampos_G No sabía que el país estaba estable...

[Alejandro Espinosa @alejandro_spi]

@ronnyMH123 @Watchdog_MP “Pay taxes for five years but don’t use the services.”

Amazing policy thinking. No private healthcare alternative, no realistic plan, just resentment and zero understanding of how the system actually works.

[Alejandro Espinosa @alejandro_spi]

@throwaway19832 @Watchdog_MP Oh look, another brave hero defending Canada from behind a keyboard and a throwaway account.

Very courageous.

[Alejandro Espinosa @alejandro_spi]

@thegrimwulff @themostbanned @DegenApeDev @Watchdog_MP Relax, hero.

You’re not saving Canada. You’re typing on X, pretending your personal frustration is patriotism.

Blaming immigrants won’t fix whatever went wrong in your own life.

[Alejandro Espinosa @alejandro_spi]

@thegrimwulff @Watchdog_MP Thank you for check my bio! God bless you!

[Alejandro Espinosa @alejandro_spi]

@DegenApeDev https://t.co/UZUdk7VT8v

[Alejandro Espinosa @alejandro_spi]

@thegrimwulff @WSB_redditor @Watchdog_MP You wrote a whole bloodline speech and still don’t know the law of your own country.
Having a child in Canada doesn’t give the parents citizenship, PR, or any automatic right to stay.
That “anchor baby” comment only exposed your ignorance.

[Alejandro Espinosa @alejandro_spi]

@m_ebrard Qué gusto ver que @diego_ruzzarin seguramente hizo esta colaboración gratis, porque cobrar por promover eventos sería demasiado capitalista.

[Alejandro Espinosa @alejandro_spi]

@DavidOndrej1 8 hours ago you were preaching DeepSeek.

Now $200/mo ChatGPT is suddenly “insane value.”

That pivot was faster than the model response time.

[Alejandro Espinosa @alejandro_spi]

@thegrimwulff @Watchdog_MP Funny how my taxes belong here, but apparently I don’t.

I’m here legally, I work, pay taxes, and support my family.

If your issue is with government policy, argue policy. Telling people to go back to “their own country” is just xenophobia dressed up as concern for taxpayers.

[Alejandro Espinosa @alejandro_spi]

@notyourc0mrade @Watchdog_MP My son is Canadian (life happens) but send my your email and I'll happy give you the money I don't need it.

[Alejandro Espinosa @alejandro_spi]

Claude Opus 4.7 is doing an excellent job.
The capability is clearly there.

My concern is token usage: it’s extremely high.
On a single serious task, it can burn through the usage limit very quickly.

That creates an unusual bottleneck:
development becomes slower not because the model is less capable, but because the cost of using it at full depth is hard to sustain.

In practice, the limiting factor stops being intelligence and starts being token efficiency.

[Alejandro Espinosa @alejandro_spi]

@z3ratul163071 @NousResearch I ran into the same thing this week. For me, it was Honcho memory. Switching from hybrid to tools fixed it.

[Alejandro Espinosa @alejandro_spi]

Not science fiction. Just a scam.

[marcus @marcusyul]

😳 LE DIERON $200 A CLAUDE Y ACCESO A UN ORDENADOR

¿El resultado?
En 10 horas pasó de $200 a $3.000 en Polymarket.

No es ciencia ficción. Está pasando ahora. 👇🏼

[Alejandro Espinosa @alejandro_spi]

Brutally true. And let's call it out: every YouTuber claiming they're "building an empire" on a Mac Studio or an NVIDIA DGX Spark with local models is lying. The only thing they're building is their YouTube channel and their X following. Nothing real, nothing that ships. @AlexFinn

[BridgeMind @bridgemindai]

Spent the last few days vibe coding on my NVIDIA DGX Spark.

Here's what I learned.

Qwen 3.5 122B took one minute and nine seconds to respond "Hi how are you doing".

Unusable for vibe coding.

Gemma 4 was fast but built a dot instead of a first person shooter game.

GPT-OSS 120B was the sweet spot.

Fast, capable, and actually produced working HTML.

Open source models running locally are not replacing Claude Opus 4.6 or Codex with GPT 5.4.

Not even close.

But they're getting better every month.

The new DGX Spark Bench is live on https://t.co/ttnnwBYerW.

Real-world benchmarks for local models on local hardware.

This is just the start. Full video below.

[Alejandro Espinosa @alejandro_spi]

Brutally true. And let's call it out: every YouTuber claiming they're "building an empire" on a Mac Studio or an NVIDIA DGX Spark with local models is lying. The only thing they're building is their YouTube channel and their X following. Nothing real, nothing that ships. @AlexFinn

[BridgeMind @bridgemindai]

Spent the last few days vibe coding on my NVIDIA DGX Spark.

Here's what I learned.

Qwen 3.5 122B took one minute and nine seconds to respond "Hi how are you doing".

Unusable for vibe coding.

Gemma 4 was fast but built a dot instead of a first person shooter game.

GPT-OSS 120B was the sweet spot.

Fast, capable, and actually produced working HTML.

Open source models running locally are not replacing Claude Opus 4.6 or Codex with GPT 5.4.

Not even close.

But they're getting better every month.

The new DGX Spark Bench is live on https://t.co/ttnnwBYerW.

Real-world benchmarks for local models on local hardware.

This is just the start. Full video below.

[Alejandro Espinosa @alejandro_spi]

@melvynx I've been running Hermes with ChatGPT across three projects and still haven't hit the weekly cap. So far, so good.

[Alejandro Espinosa @alejandro_spi]

@0x_Discover Lost my job, rent due, no backup plan — but still paying $20/month for Claude. Now that's brand loyalty.

[Alejandro Espinosa @alejandro_spi]

@adiix_official $1.18 to $3.3M. Comment 'GUIDE' to receive. The Chinese engineers must be thrilled their millions are being distributed via Twitter likes.