I create content about deep learning | The DevRel and Deep Learning Dude

Hacking on LLMs, curating multimodal datasets, and writing a book about RAG.

Harpreet Sahota 🥑

i've literally been waiting for years to have an OPEN SOURCE model like qwen3-vl-embedding

i tested its multimodal retrieval capabilities on a corpus of 412 short video clips and the results literally blew my mind

here are the queries i tested:

1. a cartoon guy drinks merlot wine

i like this query because we see how it can retrieve based on semantics (a cartoon), text (the label merlot), and temporal action in the video (the cartoon guy drinks the wine mid-way through the video)

2. garfield runs out of the door

semantics are awesome here, the model knows i'm talking about the cartoon character...

3. a woman in a blue dress sleeping on a red bench

the skeptic in you might think that it's retrieving just based on the red color block in the 2/3 of the video...but notice the specific part of the query "a woman in a blue dress..." and this is only shown in 3s out of the full 10s

this is such a huge release 

and it's gonna open up SO much more for multimodal video retrieval this year

on my wishlist is natural language search of pcd datasets, who gonna ship that?

you can hack around with the model using the resources below

check out the docs: here: https://lnkd.in/gh7yfBNP

and the quickstart nbhere:  https://lnkd.in/gbinsx7p

🤖 Hacker-in-Residence @ Voxel51| 👨🏽‍💻 AI/ML Engineer | 👷🏽‍♀️ Technical Developer Advocate | Learn. Do. Write. Teach. Repeat.

for those who missed it, or were there in-person and know the trouble i was facing 😆, my ODSC webinar is now live on youtube!

in this session i discuss visual ai techniques for automotive damage detection. 

i used the cardd dataset, which is a public dataset built specifically for vehicle damage analysis and demonstrated how fiftyone helps you explore, visualize, and build models to detect dents, scratches, cracks, glass shatter, tire flats, and broken lamps.

i feel like its a super practical workshop with some insightful workflows for visual AI in automotive settings, but as the creator i might be biased.

only one way to find out for yourself, watch the webinar!

https://lnkd.in/ewfz5vZD

Open video: Mastering Visual AI with Vision-Language Models & Advanced Evaluation Techniques by Harpreet Sahota on youtube.com

Mastering Visual AI with Vision-Language Models & Advanced Evaluation Techniques by Harpreet Sahota

sometimes i look back at the shit i get done and i genuinely impress myself

how teams really build Visual AI in the wild? what's the he messy reality?  

i don't know...and maybe its because nobody’s asking practitioners and compiling their answers so we can find out the landscape actually looks like. 

so, we at Voxel51 are kickin off the...

wait for it...

 State of Visual AI survey...vey...vey

for those out ther building visual AI systems i want your hot takes

what’s the hardest part about making visual ai a reality? 

let us know in this survey

it'll take you 30 minutes and is a much better use of time than baby sitting your runs and staring at your loss curves

i can't offer much beyond my heartiest thanks, but you will get early access to the full report with benchmarks, common pain points, and where you stand within the landscape. 

here's a link: https://lnkd.in/gnsADACy

MedGemma 1.5 supports detection, but for best results, you'll need to fine-tune. 

there's a kaggle competition using the model, i created a starter notebook to give you a jump start on how to fine-tune it for detection

checkout the starter notebook and additonal resources 👇🏼

cc Omar Sanseviero Megan Risdal Logan Kilpatrick

looking forward to the next meetup in boston!

Join our in-person meetup at the Microsoft Research Lab at the Massachusetts Institute of Technology in Boston, Massachusetts on Feb 18 to hear talks from experts on cutting-edge topics across AI, ML, and computer vision. Register here: https://lnkd.in/gi4kpSMB

Talks will include:

• SNAP: Towards Segmenting Anything in Any Point Cloud - Hanhui Wang at Northeastern University

• Culturally Adaptive AI - Anku R. at Massachusetts Institute of Technology

• Data Foundations for Vision-Language-Action Models - Harpreet Sahota 🥑 at Voxel51

• Better Together: Leveraging Unpaired Multimodal Data for Stronger Unimodal Models - Sharut G. at Massachusetts Institute of Technology

• Neural Radiance Fields for Image Verification - Sage Simhon at Pasteur Labs & ISI

#computervision #ai #artificialintelligence #machinevision #machinelearning #datascience

Syed Abdul Sami Rizvi, Matthew Ebisu and 118 other attendees

View event: Feb 18 - Boston AI, ML and Computer Vision Meetup

Feb 18 - Boston AI, ML and Computer Vision Meetup

in the asu area and wanna get started with open source contributions? then don't miss this hack

On March 21, 2026, we’re hosting Agents World: Visual AI Hackathon @Arizona State University, together with Pavan Turaga 🚀 

This is a hands-on day: meet people, build with AI, eat pizza, and leave with something you created after a short workshop and a full day of hacking. 
You’ll work with the FiftyOne ecosystem and real models used in practice, like Qwen, Gemini Vision, Sharp, and more. 

If you’re an ASU student who wants to build, learn by doing, and work with real Visual AI tools, this is for you.

Registration is open:
https://lnkd.in/gBCHW5Fg

Hope to see you at ASU. I’ll be there with the amazing Voxel51 team:  Sid Mehta, Ritchie Martori, Jimmy Guerrero ! 🙌

Inspired by the amazing Hugging Face’s 2025 AI Timeline, we created one for the Fiftyone community. The FiftyOne Community Development Timeline highlights more than 131 contributions from 2025, including datasets, models, plugins, and integrations.

You can filter by type or modality and jump straight to each integration.
https://lnkd.in/ev2kpCGN

Big thanks to the entire Voxel51 community for building all of this together.
Harpreet Sahota 🥑, Daniel Gural, Paula Ramos, PhD, Jimmy Guerrero, Derek Van de Streek, Nicholas Lotz 🚀

Gonna be a great series of talks, lots of technical depth!

Chief Community Officer @ Voxel51 | Startup Advisor

Join us on Feb 18 for the Boston AI, ML and Computer Vision Meetup! Pre-registration is mandatory to clear building security - https://lnkd.in/gHtBuH49

Talks will include:

* SNAP: Towards Segmenting Anything in Any Point Cloud - Hanhui Wang at Northeastern University

* Culturally Adaptive AI - Anku R. at MIT

* Data Foundations for Vision-Language-Action Models - Harpreet Sahota 🥑 at Voxel51

* Better Together: Leveraging Unpaired Multimodal Data for Stronger Unimodal Models - Sharut G. at MIT

* Neural Radiance Fields for Image Verification - Sage Simhon at Pasteur Labs & ISI

***********

Want to build better computer vision models? FiftyOne is an open source toolkit from Voxel51 (our Meetup sponsor) that helps you curate datasets, evaluate model performance, visualize embeddings, catch annotation errors, and eliminate duplicate images—all in one place.

“pip install fiftyone” is all it takes to get started - https://docs.voxel51.com/

#computervision #ai #artificialintelligence #machinevision
#machinelearning #datascience #physicalai

I'm passionate about harnessing the transformative power of Artificial Intelligence to solve complex business challenges and create significant, measurable value. I love to lead AI innovation, building and executing strategies that integrate cutting-edge technologies like Machine Learning, Generative AI, and advanced data analytics into the core fabric of an organization.   

With over 10 years shaping data-driven strategies and solutions , my leadership philosophy centers on building, empowering, and mentoring high-performing, cross-functional teams capable of tackling ambitious AI goals. I thrive on setting a clear strategic vision for AI, developing actionable roadmaps aligned with business objectives, and fostering an organizational culture where data literacy and AI-driven innovation flourish.   

My expertise spans the full AI/ML lifecycle, encompassing Machine Learning, Deep Learning, NLP, MLOps practices, cloud AI platforms (focused on Microsoft Azure), and robust Data Strategy formulation. In roles including Global Head of Data at QIMA and Head of Enterprise Analytics at Philip Morris International (a Fortune 500 company) , I've led the design, development, and implementation of scalable AI and data solutions. My focus has always been on leveraging data foundations not just for insights, but as the critical enabler for advanced AI capabilities that drive tangible business outcomes like operational optimization and enhanced performance.   

My background in finance and risk management  provides a strong foundation for ensuring AI initiatives deliver demonstrable ROI and align with core business needs. I excel at bridging the gap between technical possibilities and business strategy, communicating complex AI concepts effectively to diverse stakeholders, from technical teams to C-suite executives, fostering alignment and driving adoption. As an instructor and published author on advanced ML , I am committed to continuous learning and staying at the forefront of AI advancements, with a particular interest in the responsible development and deployment of next-generation AI systems.

Courtlin Holt-Nguyen

I used to think responsible AI was mostly about compliance. Check the boxes, follow the guidelines, and you’re good.

But the more I work with real-world systems, the more I realize how much is left out of the checklists. The edge cases. The unintended consequences. The moments when a model makes a decision that just feels... off.

It’s easy to talk about fairness and transparency in theory. It’s a lot harder when you’re staring at a dashboard and realizing you don’t know why the model did what it did.

Sometimes I wonder if we’re moving too fast. If we’re so focused on what’s possible that we forget to ask what’s right.

We don’t have all the answers, but the teams who take the time to ask the uncomfortable questions, who are willing to slow down and look closer, are the ones who build systems people can actually trust.

Responsible AI isn't just about following the rules but rather caring enough to look for what the rules might have missed.

There’s a myth that AI is all about the smartest people in the room. The PhDs, the math prodigies, the ones who can recite every paper from memory.

But in my experience, the teams that actually deliver are the ones who know how to listen. Who aren’t afraid to ask for help. Who admit when they’re stuck.

I’ve seen brilliant models fail because nobody wanted to challenge the assumptions. I’ve seen average models succeed because the team was relentless about feedback, iteration and learning from mistakes.

It’s not about being the smartest. It’s about being the most adaptable. The most curious. The most willing to admit you don’t have all the answers.

That’s not always easy, especially in environments where everyone feels like they have something to prove.

But if I had to bet on one thing, it wouldn’t be raw talent. It would be humility. That’s what keeps teams moving forward, even when things get tough.

Your AI business case is a "Spreadsheet Hallucination." Here is why. 📉

Stop me if you have seen this specific slide in a strategy deck recently:
-- Assumption: Copilot/GPT saves everyone 2 hours a week
-- Math: 1,000 employees × 2 hours × $50/hr
Conclusion: We just "saved" $5M! The project pays for itself by Q2!

This logic is comforting. It is clean. It is also a complete fantasy.

"Time Saved" is a soft metric that rarely translates to "Cash Banked."

Why? Because of the Fractional Labor Fallacy. You cannot fire 20% of a person. If you save 10% of my time, I don’t cost the company 10% less. I just have time to answer the emails I was ignoring, or perhaps I’m just 10% less stressed (which is great for culture, but your CFO can’t put it on the P&L). 🤨

If we want AI projects to survive the "Year 1 Audit," we have to stop pitching Efficiency and start pitching New Capabilities.

Three key shifts:
1) Embrace the Jevons Paradox: History shows that when you make a resource (like coding or content) cheaper, consumption skyrockets. AI doesn’t reduce the labor bill, it clears the backlog. The value isn’t "saving money," it’s "acceleration."

2) Target the "Impossible" Work: Don't use AI to write emails faster. Use it to review 100% of transaction logs for fraud instead of the 0.1% a human team can handle. That is a hard dollar value.

3) The Whale Oil Analogy: When electricity arrived, the initial pitch was "it’s cheaper than whale oil for lighting." But the real value wasn't better light, it was the refrigerator and the PC and all of our modern conveniences.

Stop trying to sell management on cheaper candles. 
Sell them the refrigerator. 💡

TL;DR: If your ROI calculation relies on "hours saved multiplied by hourly rate," you are presenting a fantasy. The true value of AI is not in doing t

Open article: The AI Efficiency Mirage - Why Your AI Business Case Is A Fantasy by Courtlin Holt-Nguyen • 5 min read

The AI Efficiency Mirage - Why Your AI Business Case Is A Fantasy

If your upskilling strategy relies on "passion" (read: weekends), you’re filtering out your best people. 📉

In the current AI boom, there is a silent expectation in many engineering orgs: "If you really cared, you’d be reading those arXiv papers on Saturday morning."

This "After-Hours Tax" is not just lazy management, it’s actively destroying your Senior talent pipeline.

Here is the uncomfortable reality:
👶 The Junior Dev (fresh out of school, fewer obligations) has the bandwidth to hack on LLMs until 2:00 AM. 
👩‍💻 The Principal Architect (mortgage, kids, aging parents) does not.

When you force innovation into the margins of personal life, you inadvertently design an organization where technical direction is set by those with the most free time, not the most experience.

The result? Resume-Driven Development. Shiny new tools implemented without the wisdom of failure modes because the people who understand distributed systems were too busy being parents to learn the new syntax on a Sunday. 

We need to stop treating innovation like a hobby. We need to implement a "Friday Afternoon Lock," give our senior engineers time for Learning Sprints and treat upskilling like a high priority Jira ticket. 🎫

You pay your Senior Staff for their brains, not their typing speed. Stop depreciating your own assets by refusing to buy them the time to recharge and upgrade.

undefined undefined

Executive Data Science Specialization

Udacity Azure ML Engineer Nanodegree

Code Free Data Science (KNIME)

Managing Data Analysis

A Crash Course in Data Science

Building a Data Science Team

Data Science Foundations - Level 1

IBM Blockchain Essentials

Microsoft Certified: Azure AI Fundamentals

Microsoft Certified: Data Analyst Associate

IBM Blockchain Essentials V2

Defined and executed the global AI/ML strategy for QIMA (across 100+ countries), developing a roadmap focused on leveraging advanced analytics and machine learning to drive operational efficiency and deliver unique insights that align AI initiatives with core business objectives.   

Led, mentored, and scaled a global team comprising Data Scientists, ML Engineers, BI Analysts, and Data Engineers across multiple countries, fostering a culture of innovation, collaboration, and continuous learning to accelerate the delivery of data-driven solutions. 
  
Directed the end-to-end development and deployment of machine learning models and AI-driven features integrated into QIMA's core platforms, enhancing capabilities in areas such as intelligent document analysis and neural machine language translation while accelerating existing workflows with generative AI. 

Championed and began implementing MLOps best practices to streamline the AI development lifecycle, aiming to improve model deployment frequency, monitoring, and scalability using cloud-based infrastructure. 

Delivered actionable, AI-powered insights and data solutions that directly contributed to optimizing business operations, enhancing performance measurement, and supporting strategic decision-making across the organization.   

Collaborated extensively with C-suite executives, Product Management, Software Engineering and Cyber Security leaders to define AI requirements, secure buy-in for strategic data initiatives, and ensure the seamless integration of AI capabilities into business processes and client-facing offerings.   

Initiated the development of data governance within the global data team to support self-service BI and the development of transparent AI/ML models.

Head of AI and Data

📊 Designed, built, and implemented client solutions related to data strategy & data project prioritization, predictive analytics, and data visualization for clients in North America.

🤝 Worked closely with the Vice President of Analytics at CGI to create the data strategy for one of Canada’s largest provincial governments. The project had a 5-year budget of $40 million to develop a Data Management Office for the province. Data Quality, Data Literacy, Cybersecurity, and Organizational Change Management were my primary focus areas. 

🖥️ Co-led an Agile analytics team with CGI to design, build, and implement an analytics system using R Shiny web applications for one of British Columbia’s public health departments.

Director Data Science

🎓Designed and delivered live teaching sessions for students enrolled in upGrad's professional data science and business analytics certificate program in partnership with the University of Maryland Robert H. Smith School of Business.

📊 Topics taught included Python, Deep Learning, Convolutional Neural Networks, NLP, Text Analytics, Exploratory Data Analysis, and Statistics.

🗣️ Feedback from students:
     "Courtlin was excellent in his presentation and shared his experience well."
     "Loved the session. One of the best sessions so far."

UpGrad Expert - Advanced Machine Learning, Data Visualization and Statistics Instructor

📈 Led the data-driven decision-making transformation as head of the Enterprise Analytics department (reporting directly to the Finance Director for Indochina) while promoting the use of self-service analytics.

🧑‍🤝‍🧑 Recruited and supervised a team of junior, senior, and manager-level analysts while guiding their professional development.

📋 Developed comprehensive standard operating procedures for the analytics division to improve the consistency of operations and assist with new hire onboarding.

🛠️ Managed the project team that built the company’s business intelligence system (covering its sales, distribution, and operational functions) using SQL, Python, R, and Microsoft Power BI. Created the data model design and overall data architecture for the BI system.

🤝 Operated as the primary analytics business partner to support internal stakeholders with comprehensive analyses of issues, including the performance of new product launches, geospatial analysis for distribution routes, customer segmentation, and customer profitability (ROI).

🎯 Developed an improved sales forecasting methodology using Python that was more accurate than the company’s preexisting, manually derived sales forecasts. (Forecast error was reduced by approximately 67%)

⏱️ Reduced the time spent by analysts in the Sales Development, Distribution, and Marketing departments gathering data and preparing routine management reports by 10-20 hours per month per department by measuring report usage, eliminating unnecessary reports, and leveraging Power BI automation. The time saved was used to conduct deeper, more insightful analysis of the underlying data.

👥 Led the development of a B2B customer segmentation model using unsupervised clustering algorithms that challenged several long-held company beliefs regarding which customer segments were growing and where they were located. The results were used to generate various B2B customer personas to improve trade promotion program design.

Director of Enterprise Analytics Vietnam

Data Scientist

💼 Provided internal business plan consulting, financial modeling, business process optimization, competitive marketing research & business intelligence to multiple subsidiaries within our group of companies in Vietnam.

🏆Accomplishments 

🔍 Designing, overseeing, and analyzing the competitor, market, and consumer intelligence data collected by the department’s research analysts to generate actionable insights and strategic recommendations for the Board of Directors.
💹 Creating financial forecast models for potential acquisitions.
📊 Developing KPIs, dashboards, and business scorecards for subsidiaries to align their strategic and operational priorities.
📑 Conducting business feasibility studies, developing business plans, and designing sales forecasts for potential new ventures.
🤝 Structuring and negotiating partnership agreements with counterparts from the US, EU, and South East Asia region.
📅 Assisting subsidiary companies with the creation of annual strategic plans and budgets.

* Skills Required
Business intelligence  data analysis, financial analysis, financial modeling, strategic planning

En Pointe provides core business functions including Human Resources, Digital & Information Technology, Finance & Accounting and Business Development to enable each company in our group to excel at what it does best. Our Group companies include:

1) Purpose Media 
2) Novel Production
3) Amberstone Media
4) Billboard Vietnam
5) INQ
6) Renaissance Collection
7) Embassy Education
8) AMPA Education
9) Soul Music & Performing Arts Academy - SMPAA

Director Of Business Intelligence

Kolea Consulting Group manages a portfolio of machine learning and AI-powered businesses serving customers in North America and Southeast Asia.

Director Consulting Services

Managed the firm's value-oriented investment strategy by conducting primary research on investment ideas, selecting securities, executing trades and managing portfolio risk.

Created the firm's summer intern equity investment analyst training program. Many of graduates from the training program went on to secure investment banking job offers.

Developed strategic partnerships with 3rd party financial services providers.     

Software: 
Bloomberg Professional, Factset, TD Ameritrade Institutional ThinkPipes, TD Ameritrade ThinkOrSwim, AssetBook, Morningstar

Portfolio Manager & Vice President of Business Development

Analyzed and pitched long/short equity investment ideas to the firm's Chief Investment Officer. Built financial models and valued companies on the basis of DCF, relative value and LBO methodologies. Covered companies in the technology, healthcare and industrials sectors. Executed trades and rebalanced client portfolios on behalf of the CIO.

Registered Investment Advisor (RIA)

Investment Research Analyst

Created an in-house firewall Global Management System; reduced service delivery costs by 87%.
Developed and implemented the firm’s cybersecurity assessment product offering.
Refined and documented the firm’s core operational procedures into a series of SOP manuals.
Remotely managed several hundred client PC and Mac workstations and several dozen Windows servers.
Conducted cybersecurity assessments for corporate clients, identified vulnerabilities and recommended improvements.
Designed hardware and software solutions to protect clients from malware, ransomware and spyware.

IT Systems & Cyber Security Analyst

Served as the State of Hawaii public pension fund's Investment Specialist throughout the Great Recession of 2008-2011. Assisted the Chief Investment Officer in the management of the State's $10.5 billion employee pension plan. Monitored the monthly performance and operations of the Plan's 30+ outside investment managers. Conducted reviews of several hundred prospective third-party investment managers. Created a comprehensive cash flow model for the pension fund to model various investment return scenarios. Rebalanced the Plan's asset allocation on a quarterly basis. Analyzed RFP responses for the Plan's private equity, venture capital and large-cap equity mandates.

Investment Specialist

Global Head of AI and Data @ QIMA | AI Strategy & Innovation Leader | Driving Business Value through AI/ML & Data | ex-Fortune 500 Analytics Head, AI Engineer, Hands-on developer 

Awarded for successfully managing the project team to design, build and deploy the company's business intelligence system. The project was completed on time, in full and within budget, despite disruptions from Covid-19.

Philip Morris ABCD Award - Above and Beyond the Call of Duty

English

Vietnamese

Project Management Institute

Business Networking International (BNI)

Entrepreneurs Mastermind

Honolulu Value Investing Society

Downtown Exchange Club of Honolulu

CFA Society Hawaii

Additional Organizations

Led a critical performance optimization project for an enterprise-scale AI system that processes manufacturing quality inspection reports and generates intelligent corrective action recommendations. The system leverages advanced GPT-4 models to analyze inspection findings, classify defects, eliminate duplicates, and automatically match specific defects to appropriate corrective actions from a master database. This mission-critical system processes thousands of inspection reports daily across global manufacturing operations.

Delivered breakthrough performance improvements including 60% reduction in processing time, 100% reliability under extreme concurrent load (up from 65% success rate), and 40% cost optimization through intelligent dual-model routing. The optimization eliminated all timeout-related failures and implemented bulletproof retry mechanisms, ensuring guaranteed success rates for mission-critical quality data that directly impacts product safety, regulatory compliance, and customer satisfaction across global supply chains.

GenAI summarization

Implementation of custom shape maps in Power BI to show the breakdown of Vietnam’s 2022 population by province

Custom Shape Maps in Power BI 

In this project, I created two classification models to predict whether a patient is likely to have a stroke or not.

The first classification model uses Azure AutoML to test various classifications models and selects the best one based on overall Accuracy. The second classification model uses logistic regression via sklearn and Hyperdrive to test various combinations of hyperparameters, - C (Inverse of regularization strength) and max_iter (Maximum number of iterations taken to converge), in an attempt to find the optimal model. Once the best AutoML model was found, it was deployed as a webservice and test patient data was sent to the deployed model to generate a response prediction, i.e. True or False.

Capstone Azure ML Engineer Udacity Nanodegree Azure AutoML and Hyperdrive

Udacity Machine Learning Engineer with Microsoft Azure Nanodegree Project 2
This is the second project in the Udacity Machine Learning Engineer with Microsoft Azure Nanodegree Program. In this project, Azure was used to configure a cloud-based machine learning production model, deploy it, and consume it. A pipeline was also created, published, and consumed.

Operationalizing Machine Learning

This project is part of the Udacity Azure ML Nanodegree. In this project, I built and optimized an Azure ML pipeline using the Python SDK, a provided Scikit-learn logistic regression model and Azure HyperDrive. This HyperDrive optimized model was then compared to an Azure AutoML run to determine which method produced the most accurate model.

Optimizing an ML Pipeline in Azure

• Led the project team that designed, built, deployed and maintained the Company's first integrated business intelligence and KPI dashboard system using Microsoft Power BI Premium and AWS.
• Created and delivered monthly stakeholder reports throughout the project to track progress against the timeline and budget.
• Instructed the project team members in business requirements gathering, gap analysis, KPI determination, data modeling, data engineering and dashboard design.

Combustible Cigarette KPI Dashboard System

Created an Excel model to forecast the probable cash flow requirements and funding status of the pension plan's assets based on various return and withdrawal scenarios.  

5 year cash flow projection and funding status for pension plan

As the President, I led the AOAO Board of Directors for The Kealani, a $30 million condominium in Waikiki, Hawaii.

TLDR: Expecting your engineering team to learn AI on nights and weekends isn't just a culture issue, it's an implicit bias that filters out your most 

The "After-Hours" Tax - Why Relying on Free Time for Upskilling Destroys Your Senior Staff

Agentic control is becoming increasingly important  now that the “just make it work” era of agentic AI is ending.

Product Leader @AWS | Startup Investor | 2X Linkedin Top Voice for AI, Data Science, Tech, and Innovation | Quantum Computing & Web 3.0 | I build software that scales AI/ML Network infrastructure

Everything works perfectly… 
until the agent makes one wrong move.
That’s when you realize intelligence without control is just risk.

Agentic AI isn’t about giving systems total freedom, it’s about giving them safe boundaries so they can act reliably in the real world.

Here’s a simple breakdown of the 7 control mechanisms every production agent needs:

1. Human-in-the-Loop
Critical decisions still pass through humans.
High-impact actions like payments, deletions, or irreversible changes always require approval.

2. Approval Gates
Structured checkpoints stop agents from moving ahead blindly.
Useful in multi-step workflows where conditions must be validated first.

3. Budget Limits
Agents operate within strict token, API, and compute caps.
Prevents runaway loops, unexpected bills, and uncontrolled tool usage.

4. Timeouts
Every action has a maximum execution window.
If an agent stalls, keeps retrying, or waits too long, it safely stops and falls back.

5. Confidence Thresholds
Agents only act when they’re confident enough.
Low-confidence outputs trigger escalation instead of hallucinated decisions.

6. Rollback Logic
If something fails, the system can undo it.
Reversible steps and checkpoints ensure mistakes don’t corrupt data or workflows.

7. Audit Trails
Every decision is logged and traceable.
This creates transparency, trust, and easier debugging across the entire agent workflow.

Strong agents aren’t the ones that think the most,
they’re the ones that operate inside the right boundaries.

I used to think the biggest risk in AI was technical failure. Models that didn’t converge, data that didn’t fit, code that broke in production, etc. 

But lately, I’m starting to think the real risk is something quieter. It’s building things that work perfectly...but don’t matter. 

You can optimize a model for weeks, get the metrics just right and still end up with something nobody uses. Or worse, something that nudges the business in the wrong direction because nobody stopped to ask why we were building it in the first place. 

It’s easy to get lost in the technical weeds. There’s a comfort in it, honestly. The numbers are clear, the feedback is immediate. 

But the hard part is stepping back and asking the uncomfortable questions. Who is this for? What changes if we get it right? What if we’re solving the wrong problem? 

The most important AI work happens before the first line of code. But often, that’s the part we skip in our rush to get to the fun part of the project.

There’s this moment in every AI project where the optimism starts to fade. 

You know the one. The POC worked, the demo wowed the execs, and then... reality. Integrations stall. Data quality issues crawl out of the woodwork. Suddenly, the model’s performance drops and everyone’s looking at you like you promised magic.

It's easy to get frustrated, but you should expect it to happen. It’s not a failure, it’s just the part nobody puts in the vendor case studies or in the trending YT tutorials. The part where you realize that operationalizing AI is less about the algorithm and more about the plumbing. The politics. The patience.

You can wish for a clean, linear process but that’s not how it works. There are setbacks, and sometimes you have to admit you don’t have all the answers. That’s uncomfortable, especially when you’re supposed to be the expert.

But I’ve learned that the teams who stick with it, who keep asking the hard questions and don’t flinch when things get messy, are the ones who actually deliver value. Not overnight. Not always in the way they expected.

AI isn’t a straight line. It’s a series of detours, and the best teams learn to navigate the uncertainty together.

"Magenta Coding" (that flush of AI-generated text in your git diffs) isn’t a crime. But let’s be honest, it is high-interest credit card debt. 💳

We are currently buying unprecedented velocity today by promising to pay with "understanding" tomorrow. The problem? If you max out that credit card without a repayment plan, technical bankruptcy is inevitable.

Vibe coding feels like a cheat code in the sprint but creates a dangerous knowledge vacuum in production with a dramatic increase in "orphaned code",  software that functions perfectly (for now) but is understood by no living human on the team. When the API rate limits change or the edge case hits, the "Lead Engineer" (the LLM) has lost context, and the human "Assistant" is left staring at a black box.

There are a few governance fixes you can implement now, including:
1) The "Explain It to Me" Merge Policy (No hiding behind the bot).
2) Segregating "Green Zone" prototypes from "Red Zone" core logic.
3) Documentation-First Prompting to force human architecture before AI implementation.

Don't let your developers become passengers in their own codebase. Make them take the wheel.

TLDR: The "cheat code" of AI-generated development feels like magic in the sprint, but it creates a dangerous knowledge vacuum in production. It is ti

Open article: The Magenta Mirage - Why "Vibe Coding" Is High-Interest Technical Debt for Your Enterprise by Courtlin Holt-Nguyen • 5 min read

The Magenta Mirage - Why "Vibe Coding" Is High-Interest Technical Debt for Your Enterprise

It's so easy to kee getting swept up in the hype. The new frameworks, the latest benchmarks, the endless parade of AI breakthroughs that "change everthing...again!". 

It’s exciting, sure. But it’s also exhausting. There’s this unspoken pressure to keep up, to always be on the edge of what’s next. 

But the truth is, most of the value comes from doing the boring stuff well. Cleaning data. Documenting processes. Making sure your models don’t break when nobody’s looking. 

It’s not glamorous. It doesn’t get you on stage at conferences. But it’s what actually moves the needle. 

It’s okay to slow down, to focus on the fundamentals, even when everyone else seems to be sprinting ahead.

“Emojis in print statements” is the hallmark of vibe coded work; just like “In today’s world…” is the tell for AI-generated writing.

No BS AI/ML Content | ML Engineer with a Plot Twist 🥷100M+ Views 📝

The most valuable developer skill in 2026?

Unfucking vibe-coded projects.

Here's what I'm seeing:

→ Founders ship AI-generated MVPs
→ The app works (sort of)
→ They get users, funding, traction
→ Then they try to add features
→ Everything breaks
→ They call a real developer

The codebase:
→ 1,227 branches, none merged
→ All logic in one file
→ Emojis in print statements
→ No tests. No architecture. No separation of concerns.
→ "It compiled once" is the only QA.

This is the new greenfield project.

You're not building from scratch anymore.
You're inheriting AI slop and making it production-ready.

Vibe coding creates the mess. 
Real engineering cleans it up. 

Position yourself as the cleanup crew. 

That's where the money is moving. 

💾 Save for when they ask you to "just fix" their AI prototype.
♻️ Repost if you've ever opened a codebase and immediately closed your laptop.

AI isn’t magic. You have to get through the boring part of the data audit (and inevitable data cleanup) before you can do anything remotely meaningful with AI. Convincing stakeholders that you’re going to take time and spend resources to improve data quality is the real challenge when they’re eager to see “AI everywhere” as soon as possible.

"I'm not an AI Engineer anymore," my ex-Google friend said.

"I'm a Forward Deployment Engineer now. Made $280K last year."

"Wait, what's a Forward Deployment Engineer?"

Then he showed me this GitHub Repo.

What I saw changed how I think about AI careers:

1️⃣ You Audit Data First
Not building features.
Untangling 20-year-old corrupted schemas.
Finding the "System of Record" (usually an Excel sheet).

"If you can't untangle their data, you can't build on it."

2️⃣ You Write Glue Code
Custom parsers for proprietary formats.
VPNs bridging on-prem to cloud.
Integrations the product doesn't support.

"50% integration, 50% strategy. Not 90% model training."

3️⃣ You're a Technical Diplomat
Navigating hostile stakeholders.
Turning blockers into allies.
Lead Engineer blocks VPN access? 1-on-1 to surface his fear. 
Co-author scripts to give him ownership.

"Broken trust kills more projects than broken code."

4️⃣ You Extract Patterns
Build one-off for Customer A.
Turn into primitive.
Customer B gets it as feature.

"Every deployment becomes product."

5️⃣ You Measure Differently
Not accuracy. Not training time.
(a) Outcome value delivered
(b) Your effort per $ going DOWN

"Each deployment should make the next one easier."

6️⃣ Your Stack Expands
Python, Go, SQL.
GCP: BigQuery, Vertex AI, GKE.
Terraform, Kubernetes, Google ADK.

"Full-stack: Data + Cloud + AI + Consulting."

7️⃣ Golden Rule
You Build to Become Obsolete
Building systems that run without you.
Not staying embedded forever.

Customer 1: 8 weeks embedded
Customer 5: 2 weeks embedded
Customer 20: 3 days embedded

Same outcome. 95% less effort.

"The secret?" he said.

"AI Engineers optimize for the demo. 
FDEs optimize for revenue."

Haven't thought about my career the same since.

P.S. That $280K? Base salary. Total comp: $420K with deployment bonuses. 

He's now hiring FDEs. Can't find enough people who get it.

That GitHub repo he showed me? The complete FDE roadmap.
→ https://lnkd.in/eDEfzzMq

Will you use it?

💾 Save for when your demo works but the contract doesn't close
♻️ Repost if you've been "almost ready to deploy" for 6 months

I’ve lost count of how many times I’ve been asked to “bridge the gap” between technical teams and business leaders. It sounds simple, but it’s anything but.

You walk into a meeting and realize both sides are speaking different languages. The data scientists want to talk about model accuracy and feature engineering. The business folks want to know if this will actually move the needle, or if it’s just another science project.

Sometimes I feel like a translator but other times it’s more like being a therapist. There’s anxiety (and sometimes fear) on both sides. The technical team worries about being misunderstood. The business team worries about being left behind.

I used to think the answer was more documentation, more diagrams. But honestly, it’s more about trust. About being willing to say, “I don’t know” when you don’t know. About asking the dumb questions, even if you’re supposed to be the expert in the room.

It’s not always comfortable. But the best outcomes I’ve seen come from those awkward, honest conversations where nobody’s pretending to have it all figured out.

I’ve been thinking about MLOps lately and how it’s become this catch-all for everything we don’t want to deal with in AI. 

Monitoring, compliance, data drift, reproducibility...all the unglamorous stuff that makes the difference between a cool demo and a real product. 

Nobody gets into AI for the pipelines. But if you ignore them, everything falls apart. 

I’ve seen teams pour months into model development, only to watch their work crumble in production because nobody wanted to own the operational side. 

It’s not flashy. It’s not fun. But it’s the backbone of anything that lasts. 

We may be better off if we spent less time chasing the next breakthrough and more time making sure the basics actually work.

Chat is a bottleneck for enterprise agentic AI, dynamically-generated interfaces are the next frontier. We’ve entered the era of GenUI and “liquid software”

The chatbox era is over. Welcome to GenUI.

Enterprise AI models have grown exponentially in reasoning, but they're trapped in text—the least efficient way to communicate complex data.

Transient.AI's new whitepaper introduces *Generative UI (GenUI)*: interfaces that assemble themselves in real-time to match the agent's reasoning.

Three Problems Solved:
 - The Prose Bottleneck (Cognitive Tax from text walls)
 - Static App Rigidity (no visualization = no insight)
 - Disconnected Agency (message bubbles can't guide workflows)

The Result:
* 70% faster UI development
* Interfaces that adapt to user intent
* Real-time software construction, not just conversation

The agent doesn't just "tell" you - it builds the tool you need. #AgenticAI #CaddieAI #DAF #FinTech #CapitalMarkets #AI 

Enterprise AI built by Domain Specialists. 

https://lnkd.in/e6PGsjgd

Open article: Transient AI by transient.ai

taplio

The Battle

Harpreet Sahota 🥑

Courtlin Holt-Nguyen

Followers

Posts

Engagements

Likes

Performance ScoreHarpreet Sahota 🥑's LinkedIn Performance

Performance ScoreCourtlin Holt-Nguyen's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Harpreet Sahota 🥑

Courtlin Holt-Nguyen

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);}Harpreet Sahota 🥑's LinkedIn Performance

Performance ScoreCourtlin Holt-Nguyen's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreHarpreet Sahota 🥑's LinkedIn Performance