<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Cung Đình Uy on Medium]]></title>
        <description><![CDATA[Stories by Cung Đình Uy on Medium]]></description>
        <link>https://medium.com/@uycd?source=rss-d96515a0c5d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*9z0IMFzEab-LsaVCBeeeZQ.jpeg</url>
            <title>Stories by Cung Đình Uy on Medium</title>
            <link>https://medium.com/@uycd?source=rss-d96515a0c5d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 20:35:31 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@uycd/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Thriving as an AI Product Manager in 2025: A Survival Guide]]></title>
            <link>https://medium.com/@uycd/thriving-as-an-ai-product-manager-in-2025-a-survival-guide-ba1b200e2b85?source=rss-d96515a0c5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba1b200e2b85</guid>
            <category><![CDATA[ai-trends-2025]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-product-management]]></category>
            <dc:creator><![CDATA[Cung Đình Uy]]></dc:creator>
            <pubDate>Thu, 02 Jan 2025 06:22:36 GMT</pubDate>
            <atom:updated>2025-01-02T06:22:36.633Z</atom:updated>
            <content:encoded><![CDATA[<p>If you’ve been in the tech industry for more than five minutes, you’ve probably seen some version of this meme floating around:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-gT3Y_fAKIiAFhpsF4oDXA.jpeg" /></figure><p>This meme stings because there’s truth to it. The expectations placed on AI Product Managers are escalating faster than my caffeine intake on launch days.</p><blockquote>We’re expected to be part data scientist, part strategist, part ethicist, and part fortune-teller.</blockquote><p>One minute you’re building a simple chatbot; the next, you’re expected to master reinforcement learning to build the next generation of AI agents. It begs the question: how do we not just survive but thrive in 2025 amid these escalating demands? How do we stay relevant in this dynamic field?</p><h4><strong>Riding the Wave: 2024’s AI Breakthroughs and a Glimpse into 2025</strong></h4><p>To understand where we’re heading, we need to acknowledge the whirlwind of advancements that have recently swept through the AI landscape. 2024 was nothing short of revolutionary, setting the stage for an even more transformative 2025:</p><p><strong>Smaller, Efficient Models Challenging the Giants</strong></p><p>2024 witnessed a trend where smaller models like Phi-4 and DeepSeek-V3 challenged the dominance of larger models, proving that efficiency and performance can go hand-in-hand.</p><ul><li><strong>Implication for AI PMs:</strong> This unlocks opportunities to deploy powerful AI on devices with limited resources, making advanced AI more accessible and potentially reducing costs.</li></ul><p><strong>Google’s Veo 2 Revolutionizes Video Generation</strong></p><p>Following OpenAI’s Sora, Google’s Veo 2 impressed with its ability to generate videos that demonstrate a strong grasp of real-world physics and cinematic techniques.</p><ul><li><strong>Impact:</strong> This advancement significantly impacts content creation, advertising, virtual production, and other industries relying on visual media. AI PMs should be considering the product implications of high-quality, AI-generated video.</li></ul><p><strong>Agentic AI is on the Rise:</strong> Google has made significant strides in developing AI agents capable of complex reasoning, planning, and tool use. This suggests a future where AI can perform multi-step tasks autonomously.</p><ul><li><strong>Prediction for 2025:</strong> Agentic AI may transition from research to real-world applications. While a Google advantage is possible, OpenAI and other players are also heavily invested in this area, so it’s an open race.</li><li><strong>The Evolving Role of AI PMs:</strong> These advancements are fundamentally changing the role of AI product managers. We are expected to understand, anticipate, and adapt to these changes to manage the development of AI products effectively.</li></ul><h4><strong>Becoming a Full-Stack Swiss Army Knife (Without Losing Your Mind)</strong></h4><p>So, how do we navigate this increasingly complex landscape? A key to success in many fields is having a breadth of skills and being able to adapt to evolving demands. Similarly, in the rapidly changing world of AI, being a “full-stack” product manager is becoming increasingly valuable. Now, this doesn’t mean you need to be a coding ninja who can also design a flawless UI/UX and single-handedly manage a Kubernetes cluster. But it does mean expanding your skillset beyond traditional PM boundaries.</p><p>Here’s what being a “full-stack” AI PM might look like in 2025:</p><ul><li><strong>Technical Fluency (Not Necessarily Mastery):</strong> You need to be conversant in the language of AI. Understand the fundamentals of machine learning, deep learning, different AI architectures, and their limitations. You don’t need to write production-level code, but you should be able to have informed conversations with your engineering team, understand the technical feasibility of your product vision, and identify potential roadblocks.</li><li><strong>Data Literacy is Your Superpower:</strong> Become proficient in interpreting data, asking insightful questions, and collaborating effectively with data scientists. Familiarity with data visualization tools and a basic understanding of statistical analysis are essential. Remember the adage: garbage in, garbage out.</li><li><strong>Champion Ethical AI:</strong> Ethical considerations are no longer optional. We must proactively address issues like bias, fairness, transparency, and accountability in our AI products. Be a champion for responsible AI development within your organization. This is not just a nice-to-have; it’s crucial for long-term product success and societal trust.</li><li><strong>Strategic Thinking on Steroids:</strong> AI is no longer a standalone technology; it must be tightly integrated with overall business strategy. As an AI PM, you need to connect the dots between AI capabilities and business outcomes, identifying opportunities to leverage AI for competitive advantage.</li><li><strong>Communication is Key:</strong> You’ll be juggling communication between highly technical engineers and non-technical stakeholders. Tailor your communication to each audience effectively. Develop the ability to translate complex technical concepts into clear, concise business language and vice-versa.</li></ul><h4><strong>AI: Your Co-Pilot, Not Your Replacement (For Now)</strong></h4><p>Let’s address the role of AI itself. It’s tempting to either dismiss AI tools as hype or become overly reliant on them. The key is to strike a balance. Resisting these tools will make you appear out of touch, while blindly trusting their every output is a recipe for disaster.</p><p>AI tools, especially Generative AI, are becoming increasingly powerful. They can assist with tasks like:</p><ul><li><strong>Generating product ideas and specifications</strong></li><li><strong>Automating market research and competitive analysis</strong></li><li><strong>Creating initial drafts of user stories and acceptance criteria</strong></li><li><strong>Assisting with some aspects of design and prototyping</strong></li></ul><p>However, and this is crucial, we must use these tools judiciously. They are powerful aids, not replacements for our own critical thinking, creativity, strategic judgment, and domain expertise. Think of AI as your co-pilot, helping you navigate the complexities of product development, but you’re still the one ultimately in control of the flight.</p><h4><strong>Staying Sane in the AI Frenzy</strong></h4><p>The pace of change in AI is relentless. It’s easy to feel overwhelmed. Here are a few tips for staying sane:</p><ul><li><strong>Continuous Learning is Non-Negotiable:</strong> Carve out time for learning, whether it’s taking online courses, attending conferences, or simply reading industry articles and research papers. Stay hungry. Stay foolish.</li><li><strong>Build Your Network:</strong> Connect with other AI PMs, engineers, researchers, and ethicists. Share knowledge, learn from each other’s experiences, and build a support system.</li><li><strong>Focus on the Fundamentals:</strong> While it’s tempting to chase every new shiny AI trend, remember to focus on the core principles of product management: understanding your users, defining clear goals, and delivering value.</li><li><strong>Prioritize and Focus:</strong> You can’t learn everything at once. Identify the most critical skills and knowledge areas for your current role and future aspirations, and focus your learning efforts there.</li></ul><h4><strong>The Future is Bright (and a Little Bit Intimidating)</strong></h4><p>2025 is going to be a wild ride for AI Product Managers. The challenges are real, but so are the opportunities. By embracing continuous learning, expanding our skillsets, and using AI tools wisely, we can not only survive but thrive in this exciting new era.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba1b200e2b85" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond Keywords: Building a Smarter Job Recommendation Engine with AI]]></title>
            <link>https://medium.com/@uycd/beyond-keywords-building-a-smarter-job-recommendation-engine-with-ai-00d46efb38c1?source=rss-d96515a0c5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/00d46efb38c1</guid>
            <category><![CDATA[word-embeddings]]></category>
            <category><![CDATA[ai-job-matching]]></category>
            <category><![CDATA[semantic-search]]></category>
            <category><![CDATA[vector-database]]></category>
            <category><![CDATA[resume-optimization]]></category>
            <dc:creator><![CDATA[Cung Đình Uy]]></dc:creator>
            <pubDate>Thu, 26 Dec 2024 03:20:48 GMT</pubDate>
            <atom:updated>2024-12-26T03:20:48.578Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lAfT8elh8psOIF8M" /></figure><p>My friend, let’s call him Alex, recently went through a <em>frustrating</em> job hunt. He’s a <strong>brilliant</strong> project manager, with years of experience successfully leading complex initiatives.</p><p>He applied for a role that seemed like a <em>perfect fit</em> — the description matched his skills and experience almost to a T. He crafted a meticulous resume, highlighting his accomplishments.</p><p>Here’s where things went wrong. He used the title “<strong>Programme Manager</strong>” in his resume (he prefers the British English spelling). Days later, he received a rejection email.</p><p>Confused, he dug deeper and discovered the culprit: the company’s automated applicant tracking system, likely configured with US English as the default, flagged his resume as unsuitable because it didn’t contain the keyword “<strong>Program Manager</strong>.”</p><p>Alex’s experience is a stark reminder that the traditional job search process is <strong>broken</strong>. It’s often frustrating and inefficient, leaving talented individuals like him and great opportunities lost in a sea of generic listings filtered through outdated keyword systems.</p><h4>The Frustrating Reality of the Traditional Job Search</h4><p>Alex’s story isn’t unique. How many of us have felt the <em>sting of rejection</em>, not because we lacked the qualifications, but because our resumes didn’t contain the “<em>magic words</em>” a system was programmed to find?</p><blockquote><em>The problem isn’t the number of jobs available — it’s the </em><strong><em>relevance</em></strong><em>.</em></blockquote><p>Keyword-based searches are simply too limited. They fail to grasp the <em>nuanced</em> meaning behind job descriptions and resumes, often missing strong candidates whose skills are expressed differently, or in Alex’s case, <em>spelled differently</em> than what a recruiter might search for. This leads to missed opportunities for <em>both</em> job seekers and employers.</p><h4>Unlocking Job Potential with Embeddings and Vector Databases</h4><p>But what if we could move beyond simple keyword matching and instead use AI to understand the <em>semantic meaning</em> of job postings and resumes?</p><p>Imagine a system that could have recognized that “<strong>Programme Manager</strong>” and “<strong>Program Manager</strong>” are essentially the same role, just with a regional spelling difference. This is the promise of <strong>AI-powered job matching</strong> using <strong>embeddings</strong> and <strong>vector databases</strong>.</p><p>In this article, I want to take you on a journey of exploration into an important topic, <em>inspired by experiences like Alex’s</em>. Using a project I recently worked on, I want to showcase the foundation for building an <strong>intelligent job recommendation engine</strong> using these cutting-edge technologies that could prevent such oversights in the future.</p><p><strong><em>Note:</em></strong> This demonstration focuses on building the recommendation engine itself, not on an exhaustive analysis of different embedding models or vector database options.</p><h4>How It Works: Two Key Phases of the Recommendation Engine</h4><p>At its core, our system involves two key phases:</p><p><strong>Phase 1: Job Data Ingestion and Processing</strong></p><p>Recruiters submit job descriptions, which are then transformed using AI to capture their <em>essence</em>. This transformation creates what we call “<strong>embeddings</strong>.”</p><blockquote><em>You see, AI doesn’t understand words as we do. Instead, it needs </em>numerical representations<em>. Embeddings are those numerical representations, capturing the essence and meaning of text as a list of numbers.</em></blockquote><p>We store these embeddings in a <strong>vector database</strong> — a special type of database that’s incredibly efficient at comparing these embeddings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Dwx60uYfU6n2zd7N" /><figcaption>Job Data Ingestion and Processing</figcaption></figure><h4>Phase 2: Personalized Job Recommendations</h4><p>When a job seeker provides their profile, we use the same method to create embeddings of their resume. We then use the vector database to compare this embedding to the stored job postings. The system quickly identifies and ranks jobs that align with the seeker’s skills, experience, and career goals.</p><blockquote><em>This allows us to deliver a </em>tailored<em> list of recommended jobs to job seekers, not generic search results.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*bzv22FjE_-5M2lpv" /><figcaption>Personalized Job Recommendations</figcaption></figure><p><strong>Cleaning and Normalizing: Essential Steps in Data Preprocessing</strong></p><p>To bring this to life, we used publicly available datasets of LinkedIn job postings. First, we preprocessed our data to ensure <em>quality</em> and <em>consistency</em>. This included:</p><ul><li><strong>Cleaning:</strong> Addressing missing values and inconsistencies.</li><li><strong>Normalizing:</strong> Standardizing text fields (lowercase, removal of special characters).</li><li><strong>Feature Engineering:</strong> Combining key features like job title, description, and skills into a single string to capture a <em>comprehensive</em> representation of each item.</li></ul><p><strong>Finding the Match: Leveraging Vector Similarity for Job Recommendations</strong></p><p>The <em>magic</em> lies in turning these preprocessed texts into numerical embeddings. For this project, we leveraged the <strong>Voyage AI</strong> for generating these representations. This model captures the <em>semantic meaning</em> of words and phrases in each job posting and resume.</p><p>Then, using a vector database powered by <strong>ChromaDB</strong>, we store these embeddings and perform <em>fast</em> similarity searches. When a job seeker uses the platform, we create the embedding of their resume and search the vector database to find the closest matching jobs (measured using <em>cosine similarity</em>) in terms of their meaning.</p><p>Result of finding top 3 relevant jobs of a resume:</p><pre>**Job Seeker Profile:**<br><br>*   **title:** consultant<br>*   **career overview:** ten years of experience with cerner implementations as <br>a consultant and employee. experience with all phases of implementation from <br>current state workflows to conversion for inpatient and outpatient <br>hospitals/clinics. skilled in initiating and leading implementations involving rollout of multiple ambulatory clinics within a 1-2 month timeframe. skilled in implementations for inpatient hospitals. experience in designing and building multiple components of powerchart/powerchart office/enhanced view applications to include inbox/message center, schedule viewer, powerorders, powerplans, zynx auto space, caresets, charges, task lists, easyscript, mar, all profiles, powerforms, clinical notes, powernotes, dynamic documentation, bedrock, data collection worksheet, change control process, workflow process, eprescribe, and set-up of all reference and privileges. determine estimate cost for various projects.<br>*   **qualifications:** computer skills  cerner applications powerchart <br>enhanced view, powerchart office, carenet, clinical documentation, surginet <br>documentation, powernote document viewing, production support of all cerner <br>millennium applications  cerner applications tools hna user, dcptools, scd <br>editor tool, powerplans, powerorders, cvnet tool, order management tools, <br>announcement tool, reference text tool, content manager tool, cmt nomenclature <br>tool, charge viewer, cs pricing tool, cem 500 tool, core code builder tool, <br>explore menu, bedrock, message center, eprescribe, data collection <br>spreadsheets, ops view scheduler, pref main tool, priv tool, pm launch tool, <br>citrix, reflections  other applications/software windows 7, windows 8, <br>microsoft office applications, novell, sql, tcp/ip protocols, unix/aix, <br>groupwise...*<br><br>**Top 3 Job Recommendations:**<br><br>**Recommendation #1:**<br>*   **Job ID:** 3906244147<br>*   **Relevance Score:** 0.811<br>*   **Job Title:** Client Services Consultant, Claims<br>*   **Snippet:** *...responsible for cultivating positive, long-term client <br>relationships, with our healthcare facilities and health system clients... <br>electronic remittance adviceera - 835, electronic claims  837 , and denials <br>required...*<br><br>**Recommendation #2:**<br>*   **Job ID:** 3895268999<br>*   **Relevance Score:** 0.859<br>*   **Job Title:** CT Technologist<br>*   **Snippet:** *...safely performing routine ct procedures in accordance <br>with prescribed radiation safety procedures and protocols...must have computer <br>skills and dexterity required for data entry...knowledge of ct procedures and <br>technical language...*<br><br>**Recommendation #3:**<br>*   **Job ID:** 3888965960<br>*   **Relevance Score:** 0.862<br>*   **Job Title:** MarTech Expert Solutions Architect - Customer Data Platforms<br>*  **Snippet:** *...seamlessly integrating these technologies to meet customer <br>needs while leveraging acxiom products. ... demonstrate exceptional proficiency <br>in recognizing the distinct business hurdles...*</pre><p><strong>Observe the Difference: AI vs. Keywords</strong></p><p><strong>Traditional Keyword-Based Search:</strong> A keyword search based on the resume text (“ <em>cerner implementations, inpatient, outpatient, hospitals, clinical notes</em> “) would likely produce results for roles that explicitly mention these specific keywords but potentially miss the underlying skills and related fields.</p><p><strong>AI-Powered Semantic Search:</strong> Our AI understands the <em>context</em> and <em>meaning</em> behind the consultant’s experience. Notice how the top recommendations don’t just mention the exact keywords from the resume but match the underlying skills and experience and related areas:</p><ul><li><strong>Recommendation #1: “Client Services Consultant, Claims”</strong> — This role aligns with the consultant’s experience in <em>implementations</em> and <em>healthcare</em>, even though the title and some of the keywords are different. The AI recognizes the consultant’s experience in a complex technical system, client relationships, and healthcare to connect the dots in a way that a simple keyword search would not. The remittance, claims and denials is a related area that aligns with the candidate’s skills.</li><li><strong>Recommendation #2: “CT Technologist”</strong> — Although at first glance, this appears to be an odd match, the AI understands that a CT Technologist needs to have <em>computer skills</em> and <em>dexterity</em>, skills that the consultant possesses with their past experience with complex data collection systems and their use of computer tools and software. A deeper analysis is needed to truly understand the match, but the ability to find a candidate with the technical background is a good demonstration of how this technology goes beyond keyword matching.</li><li><strong>Recommendation #3: “MarTech Expert Solutions Architect — Customer Data Platforms”</strong> — This role calls for proficiency in technical products and solutions and a deep understanding of <em>data</em>, skills that are similar to those of the consultant who has previously worked with complex data systems.</li></ul><blockquote><em>Instead of keyword matching, this process surfaces highly relevant job opportunities based on a seeker’s background and goals. This eliminates wasted time and frustration by presenting tailored recommendations rather than irrelevant listings. This is </em>transformative<em>.</em></blockquote><p><strong>The Future of Job Platforms: Enhancing Search, Recruiter Tools, and More</strong></p><p>The system we’ve created is just the beginning. Embeddings and vector databases are versatile tools that can enhance many features in a job platform, such as:</p><ul><li><strong>Semantic Search:</strong> Improve job search by matching skills, even when job titles differ.</li><li><strong>Recruiter Matching:</strong> Assist recruiters in finding the best candidates from a large pool.</li><li><strong>Content Personalization:</strong> Recommend relevant articles, courses, or industry updates.</li></ul><p>How do you see embeddings and vector databases impacting the job market? Let’s discuss in the comments!</p><p>You can also delve deeper into the technical details of this project in my <a href="https://colab.research.google.com/drive/1JSis98sPkP9KaDIQobRvfjd14qdicjqv">Colab notebook</a> .</p><p><em>Originally published at </em><a href="https://www.linkedin.com/pulse/beyond-keywords-building-smarter-job-recommendation-engine-cung-dinh-pudbc"><em>https://www.linkedin.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=00d46efb38c1" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>