Vahid Jozi

Fractional CPO - Board member

Is your product organization failing to move revenue, retention, or valuation? Product not delivering against the plan you sold the board? I step in as a senior product leader without full-time exec cost or risk.

I am the Founder and Managing Partner of FractionalCPO, a product leadership and growth consulting firm that doesn't just advise, but embeds. We work with Private Equity and high-growth enterprises to transition or augment product leadership and drive revenue growth from product.

Companies die from focusing on trying to improve metrics instead of focusing on delivering customer value. They often lack the operational velocity to turn strategy into enterprise value. We don't do performative roadmaps or endless committees. We identify the bottleneck, install high-velocity systems, and execute until the metrics move. We’ve been doing this successfully for a decade, helping clients through growth, exits, and IPOs.

🌐 Website: [FractionalCPO.com]

A bit about me:

- Builder: 4x founder and former exec (exits & failures included)
- Global: Lived in 16+ countries before university
- Discipline: 2nd-degree black belt in Taekwondo & avid meditator
- Loyalty: The guy you call when you need a ride to the airport or an ear to hear
- I’m serious about your business, but not too serious as a person.

One of my core beliefs is the power of a tribe — people who show up for each other, without being asked — which inspired me to create the Helpful Podcast, where I dig into the messiness of building, failing, and the people who show up when it counts.

🎙 Podcast: [Helpful.Show]

Let’s connect and build something real.

Board Member

A Hardware enable SaaS (HESaaS) platform for video IoT serving the likes of NASA, Disney, Fedex

the 3 types of CPOs i've worked with:



1. startup CPO: the chaos wrangler

builds the plane while it's falling. does everything from user interviews to writing specs to making coffee runs.
never sleeps. lives on adrenaline and hope.

2. scale-up CPO: the process architect

turns madness into method. builds teams and frameworks faster than engineers build features.
specializes in saying "we need a system for that."

3. enterprise CPO: the alignment wizard

navigates political landscapes like a pro diplomat. translates innovation into business speak.
spends 80% of time in meetings about meetings.

which one are you? or which one do you work with?

Revenue growth through product leadership

Agent Architect | Senior Software engineer @vivun | Cursor Ambassador

I’m an AI Agent Developer and Cursor Ambassador who has been coding with AI for the past five years, starting from the early days of GitHub Copilot. As one of Cursor’s earliest users, I now run workshops across different domains and speak frequently on AI productivity, helping developers unlock the full potential of AI-powered tools.
My expertise spans the full generative AI pipeline, from fine-tuning models and optimizing prompts to running rigorous evaluations and building scalable agents.
Grounded in a strong background in test architecture, I design AI agents that balance complexity with reliability. I focus on constantly refining and optimizing these systems to push the boundaries of what modern AI software can achieve.

Shrey Shah

Cursor just added sub-agents and skills. Here's how to use them:

What are Sub-Agents?

Specialized AI assistants that handle specific tasks independently. Each runs in its own context window.

Quick Setup:

1. Find Skills & Sub-Agents:
☑ Go to Cursor Settings → Rules, Skills, and Agent section
☑ Skills are automatically imported from Claude Code and Codex
☑ Sub-agents are also imported automatically
☑ Click any skill to see its file path

2. Use a Skill:
Just ask Cursor to use it. Example:

"Use the skill creator skill to create a LinkedIn infographic skill for me please."

Cursor will use the skill and create what you need. Next time you ask for an infographic, it uses that skill automatically.

3. Create a Sub-Agent:
Ask Cursor to use the task tool:

"Use the task tool to invoke a subagent and explore the codebase to see if there's an existing way to create a LinkedIn infographic. If not, create a subagent for creating LinkedIn infographics."

Or create one manually:
☑ Go to .cursor/agents/ folder
☑ Create a new file with your agent name
☑ Add YAML frontmatter with name, description, and model
☑ You can inherit the model from the parent agent

The Difference:

Rules = Always-on project instructions
Skills = Dynamic capabilities pulled when needed (auto-imported from Claude Code)
Sub-agents = Specialized tasks with their own context window (also auto-imported)

Why This Matters:

Instead of one agent doing everything:
☑ Skills give you domain-specific knowledge on demand
☑ Sub-agents run independently with their own context
☑ Everything syncs automatically from Claude Code and Codex
☑ No extra setup needed

Pro Tip:

Cursor automatically imports skills and sub-agents from Claude Code and Codex. Just open the Rules section and they're there. Click any skill to see where it lives.

I'm Shrey Shah & I teach AI assisted coding and agents.

And if you find it helpful, click on the ♻️ reshare button to help someone else master Cursor faster.

undefined undefined

 Shrey Shah led a workshop at JSNation US 2025

Building Knowledge Graphs with LLMs

Cypher Fundamentals

Stable Diffusion: Tips, Tricks, and Techniques

Understanding and Visualizing Data with Python

AI in Connected Products (AIOT)

 AWS Cloud Practitioner - CLF-C01 (LA)

AWS Certified Solutions Architect - Associate SAA-C02

Advanced Testing Practices Using AWS DevOps Tools

Java Design Patterns: Behavioral Part 2

JavaScript: Best Practices for Data

Networking Foundations: Protocols and CLI Tools

Data Science Foundations: Fundamentals

Deep Learning: Getting Started

Learning Kubernetes

Learning Regular Expressions

NLP with Python for Machine Learning Essential Training

Learning GraphQL

Software Testing Foundations: Bug Writing and Management

AWS for DevOps: Monitoring, Metrics, and Logging

Developing CI/CD Solutions with Azure DevOps

GraphQL Essential Training

Leadership: Practical Skills

Learning Splunk

Software Architecture: From Developer to Architect

Software Testing Foundations: Test Planning

Splunk 7.x Fundamentals Part 1

Succeeding in DevOps

Selenium Webdriver with PYTHON from Scratch + Frameworks

Automate the Boring Stuff with Python Programming

Lean Management in 2020: Agile + Kanban 

Spring Boot REST API and Microservices - Spring Cloud

Understanding Docker and using it for Selenium automation

Appium - Selenium for Mobile Automation Testing

Vivun delivers the world’s most advanced AI Sales Agent—automating the critical work required to move complex deals forward. By providing expertise and generating sales-ready outputs that help every rep engage stakeholders, build value, and drive urgency, Vivun empowers sales teams to close more deals, faster.

For more information, visit vivun.com. 

Vivun delivers the world’s most advanced AI Sales Agent—automating the critical work required to move complex deals forward. By providing expertise and generating sales-ready outputs that help every rep engage stakeholders, build value, and drive urgency, Vivun empowers sales teams to close more deals, faster.

Senior AI Software Engineer 

Software engineer | AI

V0 Ambassador

Helping bridge the gap between the community and Cursor team! If you have questions feel free to send me a message!

Cursor Ambassador

● Led development of an end-to-end test automation framework using wdio, Appium, Mocha and Saucelabs for web, mobile and API testing, significantly improving test coverage and reducing regression testing time. This allowed new features to be released faster, resulting in increased customer conversion rate.
● Implemented continuous integration and delivery pipelines using GitHub Actions, enabling teams to release new versions multiple times per day. This drastically improved time-to-market and increased customer satisfaction due to more frequent updates.
● Implemented GitHub actions for automating the entire release process, including PR creation and merging, as well as confluence release documentation, reducing release time.
● Developed backend for one of  core products using Spring Boot, supporting high volume of daily transactions and driving strong revenue growth after launch.
● Created algorithms to auto-generate test cases from historical test-data, reducing manual test creation and maintenance efforts. This allowed the testing team to keep pace with aggressive development timelines, while improving test coverage and quality.
● Integrated test automation framework with Zephyr Scale to provide end-to-end traceability. This improved visibility into overall test coverage and helped identify areas needing more test focus, reducing defects.
● Implemented Splunk dashboards and alerts for real-time application performance monitoring. This enabled pre-emptive identification and resolution of issues, improving uptime and customer satisfaction.
● Provided technical leadership and mentoring on test automation best practices using Mocha, WebDriverIO and JavaScript, improving team productivity and efficiency.
● Played a pivotal role in the recruitment process for SDETs conducting interviews to match job requirements.

Senior Software Development Engineer In Test / JAVA developer

● Conducted thorough research on potential technical and non-technical solutions, testing frameworks, and emerging tools, enhancing business operations efficiency.
● Utilized the inverse pyramid of automation (Unit, Integration/Mid-Level, and Acceptance Tests) for comprehensive feature test coverage, reducing the number of test cases.
● Developed an innovative test automation framework using Cypress, Mocha, Chai, and JavaScript from scratch, improving test automation efficiency.
● Implemented Eslint and SonarQube tools to consistently maintain code quality, decreasing bugs and errors.
● Managed test data creation with various JavaScript libraries, increasing test data accuracy.
Employed Postman, Mocha, Chai, JavaScript for API testing, and JMeter for UI/API performance testing, 
● Crafted MongoDB queries in Roto3T for database validation against the UI, enhancing data accuracy.
● Scheduled test runs, builds, and CI using Azure jobs and pipelines, streamlining the testing process.
● Leveraged Azure DevOps and Visual Studio Team Foundation for agile ALM, fostering efficient communication and collaboration between development and QA teams.

Automation Engineer

● Led comprehensive testing of applications, including Blockchain, Mobile & Web apps, SQL databases, and IoT devices, ensuring product quality and enhancing user experience.
● Developed a data-driven framework for API, Web, and Mobile app testing, improving testing accuracy and speed.
● Engineered an API testing framework using Python and Pytest for IoT devices, increasing testing efficiency.
● Conducted load testing on APIs using Apache JMeter and Artillery, optimizing application performance and reliability.
● Dockerized SoapUI and integrated it within CI/CD on GitLab, enhancing the testing process's speed and accuracy.
● Performed code quality analysis using Sonarlint and SonarQube, reducing errors in the QA framework.

SDET

● Welcomed students and directed them to the relevant office/personnel.
● Carried out clerical duties such as answering phone calls, responding to emails, and preparing documents, including office correspondence, memos, and presentations.
● Coordinated and managed appointments, meetings, and the conference room schedule in order to prevent duplicate bookings.
● Sorted and distributed mail
● Maintained general office files, including job files, vendor files, and other files related to the Basketball department.
● Purchased office supplies, equipment, and furniture.
● Provided excellent customer service.
● Performed other relevant duties when needed.

Administrative Assistant

●	Developed Selenium test cases using the Hybrid framework (Data Driven Framework with POM and TestNG)
●	Validated the scripts to make sure they have been executed correctly and meets the scenario description.  
●	Reviewed the Business Requirements Document and Test Plan to carry out the testing activities. 
●	Developed Test scenarios, Test Cases & Test data and Performed Positive and Negative testing
●	Worked closely with the automation team to create plans for what features can be automated.
●	Attended Sprint Planning and Demo to review & understand uses cases, non-functional requirements & design documents for Test Case Development

QA Automation Engineer

Intern

I teach AI assisted coding and agents | Senior Software engineer @vivun | Cursor Ambassador | V0 Ambassador

I've been coding with Claude Code for over a year.

and I wanted to share all the resources that helped me cc better. 

A lot has changed:
☑ Models have evolved
☑ Context engineering approaches have shifted
☑ New features keep getting added

Most people use Claude Code as a basic agent and miss its full potential.

I've collected resources that helped me get more out of it:
☑ Official documentation & guides
☑ Video tutorials from experts
☑ Best practices & workflows
☑ MCP servers & integrations
☑ Subagents & automation hooks
☑ SDK examples & tools
☑ Community-curated skills

I've compiled everything into one single document.

It includes:
☑ Getting started guides for beginners
☑ Deep dives for advanced users
☑ Tools & extensions to enhance your workflow
☑ Real-world case studies & experience reports

If you want these resources, comment 'claude' and I will DM everything to you.

--- 
I'm Shrey Shah & I teach AI assisted coding and agents.

If this helped you master Claude Code, click the ♻️ reshare button to help someone in your network master it too.

I teach AI assisted coding and agents | Senior Software engineer @vivun | Cursor Ambassador

Your agent is forgetful. Here's how to fix it.

4 memory types every AI agent needs:

1. Short-Term Memory
☑ Stores recent chat history (within token limits)
☑ Keeps context in a single session
☑ Like a whiteboard the agent keeps scribbling on until it runs out of space
☑ Temporary and cleared after the session ends

2. Static Memory
☑ Hardcoded facts that rarely change
☑ Think: user's name, company, birthday
☑ The "profile settings" of your agent's brain
☑ Perfect for storing permanent user preferences

3. Fact Memory
☑ LLM-extracted insights that get updated over time
☑ "User lives in Berlin" gets extracted and saved
☑ Personalization that improves as the conversation evolves
☑ Enables learning from past interactions

4. Vector Memory
☑ Uses embeddings + vector DBs to retrieve past chats
☑ Like giving your agent a searchable brain
☑ Helps find relevant info even from weeks ago
☑ Perfect for semantic search across conversation history

Combine them right, and your agent gets... well, agentic.

The CoALA framework breaks this down further:

Episodic Memory: Remembers what happened, when, and outcomes. Useful for case-based reasoning.

Semantic Memory: Stores structured factual knowledge. Contains retrievable facts and concepts.

Procedural Memory: Stores "how-to" information. Steps, workflows, and decision rules.

Working Memory: Active workspace for current task processing. Combines information from different memory types.

Most agents only use short-term memory. That's why they forget your name, preferences, and past conversations.

Add the other three, and watch your agent transform from forgetful to truly intelligent.

I'm Shrey Shah & I teach AI assisted coding and agents.

And if you find it helpful, click on the ♻️ reshare button to help someone else build better agents.

li_at

JSESSIONID

Agent memory types

I think most  people underestimate how good GPT-5.2-Codex is. 

It is my go-to model for any highly complex task because, unlike Opus-4.5, it always follows rules and stays on track. 

The only downside is that it's extremely slow.

 It's going to change soon and in a few months it might replace Opus.

I built an AI Study Buddy that actually remembers everything you learn 🤯

And it adapts to your exact learning style.

Most AI tutors forget you exist the moment you close the tab.
This one doesn't.

Here's what makes it different:

↳ It remembers every topic you've discussed
↳ It tracks your progress: learning, reviewing, mastered, struggling
↳ It generates personalized quizzes based on what you know
↳ It adapts explanations to match how you learn best

The memory system is the real magic here.

I built it using Redis vector search.
Every conversation gets embedded and stored.
So when you come back days later, it picks up exactly where you left off.

The tech stack:
→ LangGraph AI for agent orchestration
→ Redis + RedisVL for vector memory storage
→ OpenAI GPT-5.2 for reasoning
→ Tavily for real-time web search
→ Streamlit for a beautiful glass-morphism UI

The agent organizes your knowledge into four types:
→ Topics you've learned
→ Notes and insights
→ Progress tracking
→ Learning preferences

Ask it to explain Langgraph today.
Come back next week and ask for a quiz.
It remembers everything.

No more starting from scratch.
No more repeating yourself.
Just continuous, personalized learning.

I've made it 100% opensource with a step-by-step tutorial.

What subject would you want an AI tutor to help you master?

If you are interested in building AI Agents, I am going to create tutorials and opensource them for free: 
 
Two simple steps to get started:
1. Follow me
2. Star the repo: https://lnkd.in/dmR-a8G6

I will add New AI Agents and RAG tutorials added every week.

P.S. I create AI Agent tutorials and opensource them for free. Your 👍 like and ♻️ repost helps keep me going.

Don't forget to follow me Shrey Shah for daily guides on AI.

Building multi-agent systems? Here's a framework for design decisions with sub-agent architecture.

Quick recap: In sub-agent (supervisor) architecture, a main agent delegates tasks to sub-agents in parallel, then returns a final response. It supports distributed development, parallel invocations, and multi-hop interactions.

Three design decisions:

1. Sync vs async execution
☑ Sync: Main agent waits for sub-agent results
☑ Best when: Main agent needs results to continue
☑ Trade-off: Simpler but blocks conversation; not ideal for strict latency requirements

☑ Async: Sub-agents run as background tasks
☑ Best when: Independent tasks users don't need to wait on
☑ Trade-off: More responsive, better for latency, but more complex

2. Tool design
☑ Option A: Tool per agent
☑ Fine-grained control over inputs/outputs per sub-agent
☑ Custom filtering per sub-agent
☑ Trade-off: More configuration, more complex

☑ Option B: Single dispatch tool
☑ One tool dispatches to any available sub-agent
☑ Used by Quad Code and Deep Agents
☑ Critical: Expose all available agents to main agent
☑ Methods: List in system prompt, enumerate as tool arguments, or use progressive disclosure with a list-agents tool

3. Context engineering
☑ Sub-agent specifications: Provide enough info so the main agent calls the right ones
☑ Key elements: Tool/agent name, description, function args/types/descriptions

☑ Customize inputs: Filter/edit message history, inject relevant state/context
☑ Example: Add expertise level (beginner/intermediate/expert) to tailor sub-agent behavior

☑ Customize outputs: Return last message vs full message history
☑ Consider: Does the main agent need the full trajectory to complete the broader task?

Why this works: Simple to understand, supports parallel invocations, enables distributed development across teams.

The key: Think carefully about these decisions when building your next multi-agent system.

I'm Shrey Shah & I share daily guides on AI.

And if you find it helpful, click on the ♻️ reshare button to help someone else build better multi-agent systems.

sub-agent-architecture

Anthropic just released free courses. Here are 12 to bookmark:

Foundations:
☑ AI Fluency: Framework & Foundations
https://lnkd.in/eGttpUpQ

☑ Claude 101
https://lnkd.in/exrKFMX7

☑ AI Fluency for educators
https://lnkd.in/eMf2z4MN

☑ AI Fluency for students
https://lnkd.in/evZ_qiGS

☑ AI Fluency for nonprofits
https://lnkd.in/e7uf_CxZ

☑ Teaching AI Fluency
https://lnkd.in/eSV9zp-n

Technical & API:
☑ Building with the Claude API
https://lnkd.in/ehyRuUZM

☑ Claude Code in Action
https://lnkd.in/ec59N2r7

☑ Introduction to Model Context Protocol
https://lnkd.in/ebMTG6Ht

☑ Model Context Protocol: Advanced Topics
https://lnkd.in/e56nR4hh

Cloud integrations:
☑ Claude with Amazon Bedrock
https://lnkd.in/e_g3AHET

☑ Claude with Google Cloud's Vertex AI
https://lnkd.in/egf4ErhU

Why this matters:
These courses cover Claude code, AI collaboration, API integration, and MCP. They're free and structured for different roles.

I'm Shrey Shah & I teach AI assisted coding.

And if you find it helpful, click on the ♻️ reshare button to help someone else prompt better.

I'm doing a workshop at AI Coding Summit 2026 about long running Coding agents.

it's going to be about architecting long running coding agents harness using claude agent SDK and Cursor. 

It's going to be focused around Lessons I learned from Building Multi Session Agents with Cursor + Claude Code SDK 

This workshop is built around one idea:

Planning is cheap. Execution is where everything breaks.

So we’ll focus on the exact workflow that keeps agents reliable across long builds.

What you’ll walk away with:

1. A planning system in Cursor 

☑ Turn “build me an app” into a real spec
☑ Define features, architecture, constraints 
☑ Create a plan the agent can follow without wandering


2. An execution harness using Claude Code SDK 

☑ Adapt the SDK architecture to run your plan 
☑ Split work into durable steps 
☑ Add checkpoints so progress doesn’t evaporate


3. Patterns for long running, multi session agents 

☑ Context management that survives across sessions 
☑ State and memory that don’t get lost 
☑ Guardrails so the agent stays consistent from start to ship



If you’re tired of restarting agents from scratch or iterating over and over again…

This is the fix.

My official #AICodingSummit badge is here: https://lnkd.in/eKx9pzrS

I’m Shrey Shah & I teach AI assisted coding.

And if you find it helpful, click on the ♻️ reshare button to help someone else build agents that don’t forget.

Everyone is a “production AI expert” on LinkedIn now.

RAG in production. Agent pipelines. Eval harnesses. Tool calling. Observability.

And somehow… the loudest advice is coming from people who’ve never shipped a single production AI system.

It’s not even malicious half the time.

It’s just copy paste content.

ChatGPT makes it unbelievably easy to sound smart.

But here’s the scary part:

A lot of people can’t tell the difference. So they follow the advice. Then they wonder why their “agent” becomes a random text generator the moment it touches real users.

Here’s a 60 second credibility filter you can use before you trust anyone’s “how to deploy agents” post:

1. The Profile Check ☑ Have they worked on anything close to production ML or software? ☑ Any role where reliability, latency, cost, security actually mattered? ☑ Any shipped product, not “I built a demo” threads?



If the profile screams “student, creator, aspiring” but the content screams “enterprise deployment”… pause.

2. The Specificity Test Real builders talk like this: ☑ Failure modes they hit ☑ Tradeoffs they made ☑ Metrics they tracked ☑ What broke at 2am ☑ What they’d do differently



Copy paste experts talk like this: ☑ “Just add evals” ☑ “Use a vector DB” ☑ “Add tools” ☑ “Deploy with Docker” ☑ “Scale with Kubernetes”

It’s all nouns. No scars.

3. The Receipts Question Ask yourself: ☑ Did they mention any constraints? (privacy, SOC2, PII, rate limits) ☑ Did they mention any non glamorous work? (prompt injection, logging, rollback) ☑ Did they explain what “good” means? (offline eval vs online metrics)



If none of that is there, it’s usually a content remix. Not a point of view.

And look… I’m not saying students shouldn’t share.

Share your learning. That’s awesome.

But there’s a difference between: “I’m experimenting with agents” vs “Here’s how to deploy agents in production”

One is honest. The other is performance.

So next time you see a “production guide”…

Don’t just like it. Verify it.

I’m Shrey Shah & I share daily guides on AI.

And if you find it helpful, click on the ♻️ reshare button to help someone else avoid fake production advice.

taplio

The Battle

Vahid Jozi

Shrey Shah

Followers

Posts

Engagements

Likes

Performance ScoreVahid Jozi's LinkedIn Performance

Performance ScoreShrey Shah's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Vahid Jozi

Shrey Shah

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);}Vahid Jozi's LinkedIn Performance

Performance ScoreShrey Shah's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreVahid Jozi's LinkedIn Performance