Founder & Lead ML Engineer @ AI Flow | Ex-Google & Bloomberg | Building AI Agents that scale | 10 years in AI & ML | Building AI Flow, an AI & ML focused consultancy that designs, builds, and scales end-to-end AI products for startups and mid-market companies.


> What we do at AI Flow

- AI consulting - understand how to make most of your data and benefit from the Agentic AI wave.
- AI Strategy & Roadmapping - cut through hype, map the fastest path to ROI
- Custom ML & Gen-AI systems - from data pipelines to fine-tuned LLMs and Agents
- Data platforms - ETL, ingestion pipelines, document understanding end to end solutions
- Embedded AI Teams - senior engineers who plug in and deliver from day 1

> Proof over promises

- 45% pipeline cost & time drop for hyper-realistic video at Metaphysic
- 65% faster ML iterations for Google internal ML teams
- Millions saved automating invoice processing for a top-5 Oil & Gas firm
- Built tools used by 200K daily users, with more than 200 request/second at Bloomberg.
- 85% RAG & LLM pipeline cost drop through data cleaning pipelines for law companies.

> Why clients trust us

“I trusted you to deliver a good result, and that’s exactly what happened.”
- Josua Vieten, CEO @ ExoMatter

“Deep technical expertise across the stack. Mihai understands business context and proposes scalable solutions.”
- John Hack, CPO @ Interflexion

“Mihai makes it easy to get projects done. He is pragmatic, ambitious, and takes initiative.”
Paul Craciunoiu, Founder @ Uplift


> Let’s talk about building the AI future

Open to B2B, remote projects. If you need a sharp, senior team to ship real AI, book a 30-min call: antonmih.ai/meet

Mihai Anton

You wouldn't hire a mechanical engineer who only used AI-generated blueprints.

You wouldn't trust a doctor who only used ChatGPT for diagnosis.

Yet many companies hire "AI engineers" who've never debugged a production system.

Tools change. Fundamentals don't.

undefined undefined

Comment

Watch it here: https://www.youtube.com/watch?v=CLuv-X2Fm_s

Machine Learning by Stanford University on Coursera

Cambridge English Advanced English

Founder & Lead ML Engineer @ AI Flow | Ex-Google & Bloomberg | Building AI systems that scale | 10 years in AI & ML

Founder & Lead ML Engineer @ AI Flow | Ex-Google & Bloomberg | Building AI systems that scale | 10 years in AI & ML | Building AI Flow, an AI & ML focused consultancy that designs, builds, and scales end-to-end AI products for startups and mid-market companies.


> What we do at AI Flow

- AI consulting - understand how to make most of your data and benefit from the AI wave.
- AI Strategy & Roadmapping - cut through hype, map the fastest path to ROI
- Custom ML & Gen-AI systems - from data pipelines to fine-tuned LLMs
- Data platforms - ETL, ingestion pipelines, document understanding end to end solutions
- Embedded AI Teams - senior engineers who plug in and deliver from day 1

> Proof over promises

- 45% pipeline cost & time drop for hyper-realistic video at Metaphysic
- 65% faster ML iterations for Google internal ML teams
- Millions saved automating invoice processing for a top-5 Oil & Gas firm
- Built tools used by 200K daily users, with more than 200 request/second at Bloomberg.
- 85% RAG & LLM pipeline cost drop through data cleaning pipelines for law companies.

> Why clients trust us

“I trusted you to deliver a good result, and that’s exactly what happened.”
- Josua Vieten, CEO @ ExoMatter

“Deep technical expertise across the stack. Mihai understands business context and proposes scalable solutions.”
- John Hack, CPO @ Interflexion

“Mihai makes it easy to get projects done. He is pragmatic, ambitious, and takes initiative.”
Paul Craciunoiu, Founder @ Uplift


> Let’s talk about building the AI future

Open to B2B, remote projects. If you need a sharp, senior team to ship real AI, book a 30-min call: antonmih.ai/meet

AI Flow | We build AI powered products | AI, Machine Learning, and Data managed services

AI Flow

Senior Software Engineer | AI consultant | Working in AI before it was cool

Resume

Created a social network for frequent travelers, where they can post experiences, see what others explore, and follow their steps. Take a look to see how it works!

IdeaList

MihaiAnton has 14 repositories available. Follow their code on GitHub.

MihaiAnton - GitHub

Founder & Lead ML Engineer @ AI Flow | Ex-Google & Bloomberg | Building Agentic AI & ML systems that scale | 10 years in AI & ML

AI Flow #2 is up.

This week: how we helped a client cut spending from $10K/month to $2K/month, without touching the model.

After analyzing the situation, we found the problem was the input.

They were feeding entire HTML pages to the LLM (cookie banners, navigation menus, scripts, SEO content). The model was drowning in noise.

The fix was simpler than they expected. I break down what we did and the results in this week's edition.

Every time a new model drops, I get the same wave of messages from clients: should we rebuild? Should we switch? Are we falling behind?

This week it's GPT-5.4. Next month it'll be something else.

The honest answer is always the same: it depends on what you're optimizing for, and most teams haven't written that down anywhere.

- If cost is the constraint and the new model delivers similar quality for less, run the eval and test it.
- If accuracy drives revenue or directly reduces support volume, measure whether the delta is real enough to justify the migration work.
- If you're upgrading because it's newer and that feels like the right move, that's the one reason I'd push back on.

The things that determine whether your system works didn't change this week. Your data quality didn't change. Your evaluation criteria didn't change. The problem your users need solved didn't change.

A better model sitting on top of a broken pipeline is still a broken pipeline. Test it against the metrics you already track. If they move, ship it. If they don't, look at other areas that actually move them, like the data sources feeding the system.

AI Flow #3 is up.

This week: a client came in ready to invest in AI. Budget approved, timeline set.
We told them: not yet.

After analyzing their setup, we found data in 6 different systems. No tagging, no structure, half of it contradicting the other half.

Building on that foundation would have guaranteed a broken system. So, we made a different plan instead.

"How fast can we see ROI from AI?"

Depends on what you're measuring.


- Time savings? Weeks. We automated proposal generation for a construction company. From 4 hours per proposal to 20 minutes, they felt the ROI immediately.

- Revenue increase? Months. You need time to test, iterate, and prove the feature changes behavior.

- Cost reduction? Somewhere in between.


Most companies expect all three at once, immediately.

My advice: focus on time savings first, as it’s fastest to prove.

"Should we rebuild for the new model?"

The problem isn't the question itself. It's what comes next, as most teams usually ask, "Is it better?" instead of "Is it better for what we need?"

The answer depends on what you're optimizing for. Here are 3 examples.

Should we use the latest model?

2026 isn't about chasing what's new. 

It's about building systems that work regardless of which model is trending.

AI Flow #4 is up. This week: how fast can you actually see ROI from AI?

I’m sharing one case we had with AI Flow: We helped a construction company cut proposal time from 4 hours to 20 minutes; they felt the ROI in week one.

But that's time savings. Revenue increase? Cost reduction? Each has a different timeline.

The problem is that most companies expect all three immediately, and that's why projects stall before showing value.

“Can we automate this?” is one of the most common questions I get.

Sure, we can. But that’s not the first thing you should ask.

Before you jump into workflow scripts or n8n hacks, zoom out:

- Does this process actually need to exist?
- How much time or cost will automation really save?
- Do you have clarity on the pain points you’re trying to remove?

Most teams skip that step. That’s how you end up automating messy processes, which only automate more mess.

"We need AI to analyze our customer feedback."

This request came from a head of product at a 200-person company.

What they really had:

- Feedback in 7 different places
- Support tickets in Zendesk
- Feature requests in Productboard
- Sales calls in Gong
- User interviews in Dovetail
- Social mentions everywhere
- NPS surveys in their own tool
- Churn interviews in random Google Docs

Now, they wanted AI to magically connect all of it.

And here's what we told them: AI can't fix a structural problem.

First step: pick one source, the one with the highest signal (for them, it was support tickets because every issue was already tagged and categorized).

We built an AI analysis for that one source, and it worked. From that, they got insights they could act on.

Now they're adding the others. One at a time.

If you have data everywhere, don't start by connecting everything. Start by getting value from one source first.

Client last month: "We just need you to make our chatbot smarter."

Turns out their chatbot was giving wrong answers 40% of the time.

We dug deeper, and we found out that the problem wasn't the AI, but their knowledge base:

- Outdated pricing in 12 different PDFs
- Product specs scattered across Slack threads
- Support answers that contradicted each other

No model can fix messy data. It just makes confident mistakes faster.

What we did next was to spend a couple of weeks organizing their data before touching the AI. Now the chatbot works.

If your data lives in 10 different places with no single source of truth, start there.

A founder told me they spent 6 months building the perfect AI architecture, while never talking to a single user.

They launched last month, and it turns out users needed something quite different. 

Perfect systems built for the wrong problem are just expensive mistakes. I recommend prioritizing talking to users as much as possible to ensure you build what they really need.

A16z recently led a $30M round into Lio, a startup automating enterprise procurement with AI agents.

Sourcing, contract negotiation, spend management. All of it.

I'm not surprised by the funding. Procurement is one of the most automation-ready functions in a mid-size company, and one of the most overlooked.

When I talk to operations leaders about where AI could save them time, they usually mention customer-facing workflows first. Support, sales, marketing. Procurement comes up later, almost as an afterthought.

But when we actually map where decisions are slow and errors are expensive, procurement keeps appearing: requests sitting in inboxes waiting for approval, vendor comparisons done manually in spreadsheets, contracts reviewed line by line every single time.

The pattern underneath all of it is the same: high volume, repetitive structure, clear rules for what good looks like. That's exactly where automation compounds.

The question worth asking isn't whether procurement can be automated. It's which part of your procurement workflow is costing you the most time right now, and whether the process is clean enough to hand to an agent.

If the answer to the second part is no, that's where to start.

Most enterprise clients are not ready to build AI. This would be the wrong first move, anyway.

Very often, they are excited about building AI. Budget approved, timeline set, excited to start building.

Then I look at their setup. Data scattered across 6 systems with no structure. records contradicting each other, no single source of truth.

In such cases, AI won’t be a positive addition. It will only automate the chaos.

So I get back to them with a different plan. Spend two months organizing the data first, establish structure, then build AI that actually works. 

This approach might feel slow when you're excited to start. But it’s faster than building something that breaks and spending the next 6 months trying to fix it.

If everyone you talk to says yes immediately, you're probably not getting honest advice.

AI Flow #5 is up.

This week, I'm exploring the gap between demo and production, using a client situation as an example.

They had a recommendation engine already working in production. Users were getting suggestions, and the feature was live. But it performed at maybe 70% of where it could be.

After digging into the case, I noticed some problems in how it was built. It was rigid, with hard-coded paths, no observability, and every new requirement broke previous flows.

"We built a demo that works great. Now what?"

Now you find what breaks in production. 
What you should prioritize:

- Add monitoring
- Add fallback logic
- Test with real users, not perfect test cases.

Keep in mind that the gap between demo and production is where most projects fail.

AI Flow #6 is up.

This week, I'm looking at procurement, one of the most automation-ready functions in most businesses, and the one that rarely makes it onto the AI priority list.

When we map where decisions are slow and errors are expensive across the companies I work with, procurement keeps appearing. Vendor comparisons rebuilt from scratch every time. Purchase requests waiting three days for a two-minute approval. Contracts reviewed line by line for suppliers the company has worked with for years.

Nobody designed it this way. It just accumulated.

This edition covers why this happens, what to look for in your own workflows, and where to start if the process isn't clean enough to automate yet.

xpath.global

xpath.global- a SaaS enabled Marketplace created by professionals for professionals: founded by passionate HR Executives, online market experts and global mobility specialists, xpath.global entered the global mobility market to embody a vision of an easier, faster and more transparent global mobility industry and to meet the growing need of clients and providers of fast and seamless connection.
Through cutting-edge technology and ground-breaking tools and features, the xpath.global SaaS enabled marketplace connects thousands of clients with thousands of vetted service providers around the world, as we make our main objective to offer various specialised global mobility services, anytime, anywhere, on any device.

Chief Operations Officer

Nestlers Group

Assisting clients and employees on Immigration procedures applicable in Romania and the EU;
Developing efficient operational plans on immigration, labor law and social security on behalf of the clients;
Planning the monthly immigration and labor laws trainings for the local team;
Developing Global Mobility Policies for various clients;
In charge with global mobility projects covering immigration issues, labor law and social security issues; 
Responsible  to recruit and manage staff, including performance monitoring, mentoring and training;
Responsible with the client relationship and the effective communication with the Public Authorities

Founder&Senior Immigration, Labor law and social security Partner

ELECTROMAGNETICA

Coordinated the work of the Human Resources staff 
Responsible for recruitment, interviewing, hiring and onboarding processes 
Developing and executing the performance management plan annually  
Developed and implemented the KPIs in the production unit, R&D departments and Administrative Departments
Developed job descriptions for the production unit and the R&D department
Ensured compliance with national employment law 
Analyzing the comp&ben policies
Responsable with the strategy of personnel fluctuation and retention 
Developed various human resources plans and procedures for all company personnel
Planed, organized and controlled all activities of the department
Participates in developing goals, objectives and systems for the Human Resources Team
Develops team building and training programs for the team 
Negotiating with the Company's Union the Collective Contract and coordinating the implementation and developing various comp&ben policies within the internal CCM of the company

Human Resources Director

Pirgroup Romania

Provide support for clients processes including onboarding paperwork, HR system data management, immigration documentation and cost estimate data.
Planning for clients the global mobility services developing structured relocation plans, immigration, housing, settling-in services,  income tax, cross cultural training, language training, education consultation, etc.
Stay informed about & take action on changes in local country immigration legislation, 
Participate in broader mobility projects on an ad-hoc basis within the other branch offices of the group.
Representing the clients in front of local external authorities for the global mobility area.  Keeping clients informed of local changes and best practices in the global mobility area.
Analyze case documents and prepare immigration applications for filing with the appropriate government agency.
Drafting and coordinating petitions, applications and specialized support letters detailing eligibility for visa classification and other immigration benefits. 
Representing the company in Romania.

Managing Partner

I like to think about 𝐦𝐲 𝐥𝐢𝐟𝐞 𝐬𝐭𝐨𝐫𝐲 as a puzzle 🧩 that has been completed over the years. All my experiences brought me where I am today.
 
👉 𝐄𝐗𝐏𝐀𝐓 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞: My story began with getting a scholarship to study in Belgium, offered by the Belgian Government. Back then, Romania wasn’t part of the European Union, so I had to learn immigration by doing it myself and managed to take care of the visas, permits and relo. Kind of DIY project. 😅
 
👉 𝐇𝐑 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞: My journey continued working in a Dutch-based DSP company for a couple of years. There I’ve built from scratch all the departments. Given my law background, I fell in love with immigration and human mobility processes. After some years my job and my role changed and I became HR Executive for a large company listed on the stock exchange. For 3 years my role was to restructure departments and rebuild the company’s culture and branding.
 
👉 𝐁𝐔𝐒𝐈𝐍𝐄𝐒𝐒 𝐎𝐖𝐍𝐄𝐑 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞: With all these years spent in building, structuring and restructuring in different companies, I decided to build my own boutique consultancy legal firm offering compliance for the mobility industry: immigration, tax & labor law, payroll, employer of record, and relocation services. Very soon my company became one of the most awarded boutique consultancy firms in our area of expertise.
 
👉 XPATH.GLOBAL: After 8+ years of helping thousands of clients with their mobility processes and being close to every case and fulfilling personal and sensitive issues and with a 360 perspective in mind, our actual digital start-up company, xpath,global, was born.
 
⭐ For most start-up founders, the experience is built along the way, having completely different backgrounds before. I am exactly the opposite. 𝐈’𝐯𝐞 𝐛𝐞𝐞𝐧 𝐢𝐧 𝐭𝐡𝐞 𝐄𝐱𝐩𝐚𝐭, 𝐇𝐑 𝐚𝐧𝐝 𝐒𝐞𝐫𝐯𝐢𝐜𝐞 𝐏𝐫𝐨𝐯𝐢𝐝𝐞𝐫’𝐬 𝐬𝐡𝐨𝐞𝐬. 𝐈’𝐯𝐞 𝐬𝐞𝐞𝐧 𝐭𝐡𝐞𝐢𝐫 𝐬𝐭𝐫𝐮𝐠𝐠𝐥𝐞𝐬. 𝐈’𝐯𝐞 𝐥𝐢𝐯𝐞𝐝 𝐭𝐡𝐞𝐦 𝐨𝐧 𝐦𝐲 𝐨𝐰𝐧 𝐬𝐤𝐢𝐧. I know this industry like the back of my hand with all the needs, wants, and pitfalls.
 
This is why 𝐱𝐩𝐚𝐭𝐡.𝐠𝐥𝐨𝐛𝐚𝐥 𝐢𝐬 𝐭𝐡𝐞 𝐭𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐭 𝐫𝐞𝐬𝐩𝐨𝐧𝐬𝐞 𝐭𝐨 𝐭𝐨𝐝𝐚𝐲’𝐬 𝐠𝐥𝐨𝐛𝐚𝐥 𝐦𝐨𝐛𝐢𝐥𝐢𝐭𝐲 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬: a global ecosystem where the B2B part of the funnel can find the best providers all over the world for the right prices, while the expats are able to stay up to date with the progress.
 
For more details about how xpath.global can help HR Departments, Service Providers and Expatriates send me a message 📩 and let’s connect!

taplio

The Battle

Mihai Anton

Elena Antoneac GMS-T

Followers

Posts

Engagements

Likes

Performance ScoreMihai Anton's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Mihai Anton

Elena Antoneac GMS-T

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Mihai Anton's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreMihai Anton's LinkedIn Performance