Amy Chang

Whoever claims building RAG systems is easy has never built one at scale. 🤯 There is a significant gap in the literature and existing frameworks. RAG architectures are designed for POCs, not for hundreds of thousands of documents or supporting tens of thousands of users. Designing these systems requires a cross-functional team with software engineering and data science skills blended with a product mindset. Earlier this summer, I spoke at the Databricks AI Summit, highlighting guiding principles we are taking at CVS Health for building enterprise-grade RAG systems. Every company is unique, and while there is no one-size-fits-all solution, here are three principles that will help you on your journey: 👉 Spend time on discovery—It's tempting to dive in and start building immediately, but you may get lost in the desert without doing your due diligence and understanding what delivers the most ROI. Our team spent approximately a quarter on discovery. 👉 Design for modularity—The LLM space is changing rapidly, and you want to build reusable components that can easily be swapped out or upgraded when something better comes along. 👉 Humans + Machines—Technology alone can not solve the knowledge search problem. Develop processes and governance to help teams curate the data, which will be ingested into your system. These lessons are just the tip of the iceberg. Watch my full Databricks presentation and leave your comments and questions below to learn more tips and tricks. #RAGSystems #EnterpriseAI #DataEngineering #MachineLearning #AIArchitecture #AI #DataScience #LLM #SoftwareEngineering #AIAtScale #KnowledgeSearch #TechLeadership #AIFrameworks #AIInnovation #DatabricksAISummit #ProductMindset #CVSHealth

62

5 Comments

𝐃𝐚𝐭𝐚 𝐬𝐨𝐮𝐫𝐜𝐞 𝐭𝐫𝐢𝐚𝐠𝐢𝐧𝐠 - Just as patients in an ER need to be triaged to prioritize care and stabilize those in need, data science demands a similar triage approach. Not all data sources are created equal, and some might require substantial cleaning, analysis and transformations before they're usable. The 80/20 rule of data triaging means we are going to quickly determine whether a data set will be useful by examining some key factors. ●     First is Availability: Is the data public? Is it behind a paywall? Do you need special permissions and are there security concerns? Knowing major logistics barriers can prevent chasing a dead end. ●     Second is Costs: Data can be surprisingly expensive. Are there usage fees? Storage costs? Will you need to invest in new software to process it? ●     Third is Utility: Does the data actually contain the information you need? Beware of proxies - just because it's related to your topic doesn't mean it's the right fit. ●     Fourth is Frequency of Updates: If you're predicting a real-time outcome, then stale, annually updated data probably won't cut it. Find data with a time resolution and update frequency that aligns with your problem. ●     Fifth is Geographic Resolution: Studying zip code level sales trends? Nationwide data won't help. Make sure the granularity of the data matches your scope.   I always find it is useful to obtain a quick sample of the data. This quick look can often reveal formatting issues, missing values, and whether it's worth a deeper dive. Data triage is about making tough calls. Perfectly clean, well-aligned data is the unicorn of our field. Be prepared to say "no" to a source that looks initially promising but will take an inordinate amount of time to fix for minimal gain. Being resourceful is key! Don't be afraid to ask colleagues, search online forums, and get creative. Sometimes a great data source is hiding in plain sight. #data #datascience #datasources

2

𝐃𝐚𝐭𝐚 𝐬𝐨𝐮𝐫𝐜𝐞 𝐭𝐫𝐢𝐚𝐠𝐢𝐧𝐠 - Just as patients in an ER need to be triaged to prioritize care and stabilize those in need, data science demands a similar triage approach. Not all data sources are created equal, and some might require substantial cleaning, analysis and transformations before they're usable. The 80/20 rule of data triaging means we are going to quickly determine whether a data set will be useful by examining some key factors. ●     First is Availability: Is the data public? Is it behind a paywall? Do you need special permissions and are there security concerns? Knowing major logistics barriers can prevent chasing a dead end. ●     Second is Costs: Data can be surprisingly expensive. Are there usage fees? Storage costs? Will you need to invest in new software to process it? ●     Third is Utility: Does the data actually contain the information you need? Beware of proxies - just because it's related to your topic doesn't mean it's the right fit. ●     Fourth is Frequency of Updates: If you're predicting a real-time outcome, then stale, annually updated data probably won't cut it. Find data with a time resolution and update frequency that aligns with your problem. ●     Fifth is Geographic Resolution: Studying zip code level sales trends? Nationwide data won't help. Make sure the granularity of the data matches your scope.   I always find it is useful to obtain a quick sample of the data. This quick look can often reveal formatting issues, missing values, and whether it's worth a deeper dive. Data triage is about making tough calls. Perfectly clean, well-aligned data is the unicorn of our field. Be prepared to say "no" to a source that looks initially promising but will take an inordinate amount of time to fix for minimal gain. Being resourceful is key! Don't be afraid to ask colleagues, search online forums, and get creative. Sometimes a great data source is hiding in plain sight. #data #datascience #datasources

8

Some interesting commentary in this post to consider

4

Exciting news... the AI/ML team at Cedar is growing! 🌳 I'm hiring two Data Scientists to design and build AI and ML features that differentiate Cedar’s offerings and optimize our product. 🚀 Take a look at the job descriptions to learn more about each role (they're different!) and about the impact you can make here at Cedar. You can also check out my YouTube videos below (and yes, I am using these videos as an excuse to show off my favorite feline friends 😻 #PurrProgramming #PawsOnTheJob). Sound interesting? If so, please apply directly to the role that you feel you would be most suited for! I may not be able to respond to everyone in a timely manner on LinkedIn, but if you apply, then your profile will be reviewed by someone on the team. Let's reshape the future of healthcare! 💡 Data Scientist III (Artificial Intelligence) YouTube Video: https://lnkd.in/gpHCdVHA Application: https://lnkd.in/gWKCePsK Data Scientist III (ML Engineering) YouTube Video: https://lnkd.in/gMhKcXUR Application: https://lnkd.in/gNyWHNby #hiringnow #healthtech #datascience #artificialintelligence #machinelearning #growwithcedar

204

15 Comments

❓ Did you know that Tableau's character sort is case-insensitive?   ⚖️ That means that it treats capital letters and lowercase letters as the same.   👍 That's probably a good thing and a useful default.   😀 The (somewhat funny) problem that I ran into is that has my RDBMS (Teradata) has case-sensitive sorting.   📃Capital letters sort first, lower case letters later.   ❌ I was testing my dashboard with some queries against the source data and many of my metrics were off for the first row.   🔍 In the Teradata tool, the text fields were also being truncated for display as in the "Output" section of the image.   👁️ So initially, I couldn't see the sort disparity.   👍Once I saw it, I wrapped my ORDER BY with an UPPER(variable), which eased the comparison, and everything was fine.   📝 Sometimes it's the little things (assumptions) that cause problems.   🏗→🧠 Build to Learn!   💭🚶‍♀️🚶‍♂️ Follow for more.   #data #datascience #analytics #VizoftheRay

10

The CT Public Health Fellowship Program offers experiential learning opportunities with public health agencies for students across Connecticut. "Our team at DataHaven investigates data coming in from national and statewide sources on a daily basis, and we also engage Connecticut resident in our own interviews and surveys, which are always fascinating." says Mark Abraham, a preceptor for the program at DataHaven. "I love to see the different ways that this information is used by people to help make their communities healthier and happier places to live." See what current fellows are learning and how they are contributing to public health services in Connecticut: https://ow.ly/eMm550SThM8 Connecticut Department of Public Health

27

🤔 Have you accidentally closed a card, like a Title, Filter, or Color/Shape/Size Legend?   👀 And you want to see it again?   There are several ways to reopen it if you are on the worksheet: ◼ Right-click in the area below the Marks card and then select the card to display. ◼ If you are out of room because of a lot of Measure values, right-click in the open area below your graph and then select the card to display. ◼ If your graph also takes up your entire view, and you can aim carefully, right-click ANY space between cards/shelves (like between the Columns and Rows shelves). ◼ If your aim is poor, the Shelves and Marks card are available under: Worksheet > Show Cards The legends are available via: Analysis > Legends / Filters / Highlighters / Parameters   I hope that helps.   Bonus: This is an easy way to get to the Summary card, the existence of which I always forget.   🏗→🧠 Build to Learn!   💭🚶‍♀️🚶‍♂️ Follow for more.   #tableau #data #analytics #VizoftheRay

10

3 Comments

For anyone wanting to understand data so they can make sure their charity is using it effectively then take a look at this fab course!

6

2 Comments

👉🏼 Does using artificial intelligence take the person out of personal statements? We can't tell 🤓 Jenna N Whitrock 👇🏻 https://lnkd.in/eApWGiGv 🔍 Focus on data insights: - 📊 Only 59% of AI-generated personal statements were correctly identified by participants. - 👩‍⚕️ 60% of AI-generated statements were rated as the best, while 43.3% were labeled the worst. - 📝 A significant 66.7% of participants opposed the use of AI in personal statement writing. 💡 Main outcomes and implications: - 🚫 Concerns about the reliability of distinguishing AI-generated content raise issues for the integrity of residency applications. - 🤔 The mixed perceptions towards AI usage could influence future policies regarding personal statement submissions. - 🔍 Insights suggest a need for better training or guidelines for evaluators to recognize the nuances of AI-generated text. 📚 Field significance: - 🏥 Highlights ethical considerations in surgical residency applications amidst technological advancements. - 🌐 Supports ongoing discussions about transparency and authenticity in medical education assessments. - 📈 Indicates a trend toward more stringent evaluation practices as AI becomes integrated into application processes. 🗄️: [#artificialintelligence #medicaleducation #surgery #personalstatements #ethics #residencyapplications]

1

1 Comment

Somava Saha, very comprehensive and thoughtful plan for fully leveraging and sustaining the talents of CHWs. This is critical work as we more fully acknowledge the impact of HRSNs and health equity on health outcomes. Used properly, CHWs are an indispensable tool in population health equity. Thanks for you long-standing leadership in promoting the CHW role, Soma. It just makes sense.

4

1 Comment

🏃‍♂️I'm a little late to the game, but I'd like to thank IMDb.com for featuring my Breakfast Club viz among several others on their website as part of the #dataplusmovies project with Tableau. 🙏 It's an honor to have my work alongside such noted Tableau developers. Links in comments. 🏗→🧠 Build to Learn! 💭🚶‍♀️🚶‍♂️ Follow for more. #tableau #data #analytics #VizoftheRay

36

8 Comments

For those of you who have been following our work on Pathways to Population Health Equity, one beautiful outcome of applying that framework in Rhode Island has been the recognition of the full scope of the #CHW role in building bridges between clinical and community environments to create upstream investments in population and community health. Drawn from communities closest to the challenges, these natural civic connectors, who skilled at relationships and navigation but can also be trustworthy and effective leaders of clinical-community coalitions and partnerships to address upstream community conditions like housing availability and tenant rights and root causes like racial inequities in the environment. In this week's issue of NEJM Group, we share how this led to a roadmapping process among 140 CHWs and allies about how to sustain the role with braided and blended funding and describe the process. The four roadmaps developed to sustain the role of CHWs were about: 1) Growing the role, agency, and leadership of community health workers 2) Workforce development for CHWs and their supervisors 3) Payment and sustainability 4) People and place-based coordinating infrastructure The full roadmaps can be found here: https://lnkd.in/eND39AEU A year after implementation, there remained strong engagement and support for the implementation within the state and advancement of key functions to support CHWs. More remains to be done to align funding for long-term sustainability. https://lnkd.in/eXvNvsEx Special appreciation to co-authors and co-leaders in this process, Swanette Salazar Geraldine McPhee, Deborah Garneau and many others including James Day, Linda Cabral, Kristen Rego, Michelle Mooney, Michelle Abuna, Jonathan Scaccia and our whole team at Well-being and Equity (WE) in the World who supported this roadmapping process and contributed to building CHW leadership along the way, including Hilda Ortiz, Sarai Arpero at Latino Health Access and Kevin Barnett at CACHE #equity #healthequity #populationhealth #civicmuscle #vitalconditions #P2PHE Durrell Fox

58

For those of you who have been following our work on Pathways to Population Health Equity, one beautiful outcome of applying that framework in Rhode Island has been the recognition of the full scope of the #CHW role in building bridges between clinical and community environments to create upstream investments in population and community health. Drawn from communities closest to the challenges, these natural civic connectors, who skilled at relationships and navigation but can also be trustworthy and effective leaders of clinical-community coalitions and partnerships to address upstream community conditions like housing availability and tenant rights and root causes like racial inequities in the environment. In this week's issue of NEJM Group, we share how this led to a roadmapping process among 140 CHWs and allies about how to sustain the role with braided and blended funding and describe the process. The four roadmaps developed to sustain the role of CHWs were about: 1) Growing the role, agency, and leadership of community health workers 2) Workforce development for CHWs and their supervisors 3) Payment and sustainability 4) People and place-based coordinating infrastructure The full roadmaps can be found here: https://lnkd.in/eND39AEU A year after implementation, there remained strong engagement and support for the implementation within the state and advancement of key functions to support CHWs. More remains to be done to align funding for long-term sustainability. https://lnkd.in/eXvNvsEx Special appreciation to co-authors and co-leaders in this process, Swanette Salazar Geraldine McPhee, Deborah Garneau and many others including James Day, Linda Cabral, Kristen Rego, Michelle Mooney, Michelle Abuna, Jonathan Scaccia and our whole team at Well-being and Equity (WE) in the World who supported this roadmapping process and contributed to building CHW leadership along the way, including Hilda Ortiz, Sarai Arpero at Latino Health Access and Kevin Barnett at CACHE #equity #healthequity #populationhealth #civicmuscle #vitalconditions #P2PHE Durrell Fox

19

2 Comments

I contend that Polars is the future of data analysis. As I continue to explore this wonderful package, I am amazed and inspired by its alignment with the dplyr package. I delved into data filtering and selection in Part 2 of Learning Polars the Right Way. This article highlighted various ways of filtering data. Please share my work with your network and leave a comment on how to improve future articles or topics you want me to write about. #pydata #datascience #python #deeplearning #rstats

9

I'm at the Databricks AI and Data Summit in San Francisco this week. The question I am seeking to answer is "How can we make great decision-making more affordable and accessible to everyone?" Currently, it's really hard and expensive to produce great analytic insights. And those insights are super valuable. Let me explain with an illustration. How mass production has elevated humanity: ---------------------------------------------------- Over the past 300 years, the masses of humanity have elevated their quality of living through mass production. Take the example of the Green Revolution in agriculture: The introduction of high-yield crop varieties and advanced farming techniques dramatically increased food production. This helped reduce hunger and improve nutrition worldwide, showcasing the transformative power of mass production. By lowering the cost of goods and services, we’ve made them more affordable and available to more people. Why can't we translate this idea to the realm of insights?

4

Our friends at the Massachusetts Technology Collaborative recently launched the third Challenge in their Digital Health Sandbox Challenge Program, designed to support the goals of the Healey-Driscoll Administration’s Advancing Health Equity in Massachusetts (AHEM) Initiative. This year, selected startups will compete for prizes worth $250,000. Do you have a digital health-based solution that helps eliminate racial, economic, and regional disparities in health outcomes? The Massachusetts eHealth Institute (MeHI) is seeking startups and entrepreneurs to compete in the third Massachusetts Digital Health Sandbox Challenge. This Challenge is divided into four tracks, all aimed at meeting the goals of the Healey-Driscoll Administration’s Advancing Health Equity in Massachusetts (AHEM) Initiative: • Addressing Social Determinants of Health (SDOH) for Cardiometabolic Diseases; • AI for Health Equity; • Maternal and Infant Health; and • The Medicaid Pilot Track (MPT) Challenge finalists will compete for prizes worth a total of $250,000. Finalists will also take part in Mass Digital Health Connects, a three-month series of informative and interactive virtual workshops featuring prominent voices from across the ecosystem, including venture capitalists, state and federal policymakers, and leading researchers. Applications are due December 19, 2024. Learn more and apply here: https://lnkd.in/eJYS7_ZP.

4

⚠️ New Viz Alert ⚠️   🔧 I upgraded the Seasonality Reality Checker.   😲 This is my first radar chart believe it or not - mainly because don't see a ton of value in them.   ✔ However, I don't want to create a false impression as if I am limited to bars, like I used for the prior week's #MakeoverMonday.   So radars were this weekend's challenge - they are surprisingly similar to dot/circle radials and bar radials. Anyhow, the how-to came from a talk by the Flerlage twins in #data19. Link in comments.   👀 This is actually a somewhat bad use case for a radar chart, as I mention in the video - typically you'd want multiple measures at once spiraling around a central axis with one colored polygon per dimensional element being compared - like a player or a team or candidate - rather than a single measure by slices of time as in my example.   🌹Nonetheless, you can make some cool images with it. And "beauty is its own excuse for existence" - Ralph Waldo Emerson.   And with the right parameters and possible additional filters, there are meaningful use cases in the dashboard.   If you want to understand the calculations behind these types of graphs, check out my how-to section. I do three things differently than most: ◾Naming - Spoke Count and Spoke Number are clearer to me than Index. ◾% around circle approach rather than radians when dealing with Spokes. ◾For the radials bars, I use a tiny 4-cell table and a relation rather than duplicating the datasource, which could be massive. I'm not sure if that is an innovation…   Anyhow, I hope this helps you to understand radar charts and use them where applicable.   🏗→🧠 Build to Learn!   💭🚶‍♀️🚶‍♂️ Follow for more.   #tableau #data #analytics #VizoftheRay

39

7 Comments

I recently attended MeasureCamp New York and had the pleasure of listening to speakers from a variety of data backgrounds. Initially, I approached the conference with some skepticism, given that my past data conference (not Measure Camp in particular) experiences were heavily vendor-centric. Yet, I was pleasantly surprised to find a consistent emphasis on data strategy across most presentations and discussions. This common thread prompted me to reflect on a significant transformation occurring within the data industry. When I began my career at the height of the "Big Data" era, conference discussions largely revolved around leveraging an array of measurement and data tools, with an overarching goal of amassing as much data as possible. Today, the focus feels different. Rather than being gluttons for data, data conservatism is being practiced. Where previously the majority of focus in solving business problems was on data collection and exploration, it is now spent on data strategy. This change is characterized by prioritizing the right questions, identifying the appropriate stakeholders to address these questions, and utilizing data as supportive evidence for decision-making and storytelling, rather than making it the central focus. It represents a more structured and resource-efficient approach to data utilization. As I work to foster a strategy-driven culture within my team (see supporting article in the comments), it was reassuring to see my beliefs echoed at Measure Camp. I am hopeful that data strategy will continue to gain prominence in conferences throughout the year and become a strategic focus in all organizations. #data #datastrategy #analytics #measurecamp

24

3 Comments