Embeddings, Feature stores, and MLOps with Simba Khadder

Episode timestamps(02:15): Tacita's unconventional career path to becoming a CTO  (07:00): Textio's practices for building AI responsibly and ethically  (14:00) The impact of Textio's AI on performance feedback  (17:00) The importance of purpose-built vs generic AI models(28:00) Balancing open source and proprietary data/models  (42:00) Advice for the AI industry moving forward  QuotesTacita Morway“When you've got a team with different backgrounds, educational, lived experiences, identity, careers, all of those things, we have those different perspectives in the room. And we're all working off of the same expectations. We can catch each other's gaps.”Charna Parkey“There's an interesting conversation happening, I think, in the community right now about these purpose-built LLMs. Are they as good as generic LLMs? Sure, certainly if you're not going to apply something purpose-built to something generic or outside of its domain, it is not as good. But I think some of this shows us that unless you have something purpose-built and unless you're leveraging the data in the right way, you may just be feeding noise back into the system.” LinksConnect with TacitaConnect with Charna 

Timestamps(00:02:29) Fabiana's journey starting YData and becoming a public speaker (00:20:19)  Misconceptions and hype around generative AI and AGI (00:32:46) Potential real-world impact and use cases of LLMs today (00:34:55) The role of synthetic data in making AI models more robust and fair (00:43:55) Advice for founders: value your time and learn to say no (00:48:24) The importance of technical leaders being able to communicate well  QuotesCharna Parkey: "It's a balance. I think that's also what led us to some of the demographic based data science. Essentially, folks were making like event data into pre-aggregated data. And then they were trying to obscure it so much that you couldn't get back to the person. And so you're like, okay, what's their age and what's their gender? And you're like, that's not actually the most useful part of data science that can't predict behavior or intent or any of that. It throws out time as a component of the entire process, seasonality, everything. And so there just, there has to be a better way."Fabiana Clemente: "I have to say, that's a very beautiful way to put it. Hallucinations, I have to say. I never thought about that. And it makes a lot of sense. I do think, though, that in terms of LLMs, it's so language, it's so definitely, it sounds like we are getting very, very intelligent system, exactly, because language is very complex. And we know that was needed for the leap of humanity. I do think there are other, the sense of combining. Well, and here we enter in the multimodal kind of space. It's what's missing." LinksConnect with CharnaConnect with Fabiana 

Episode timestamps(02:15): Challenges of collecting open source usage data(22:06): Driving impact with open source usage data(28:27): Avi's entrepreneurial journey(39:42) Persistence and vision in startups(44:03) Tracking outcomes to stay motivated QuotesAvi Press“I mean, one thing is, for any project that you might be thinking about doing or any initiative that you want to work on or goal that you have, I think there's a lot of power in just trying the thing. You may not have all the details figured out, but just try it anyway and see where it takes you. And I think a lot of projects that I've ever worked on that led anywhere, I didn't know all these details, but I just start trying and seeing what works anyway and being very open to it not working out, but attempting it anyway. And then the other thing, which is I think admittedly fitting into our agenda at Scarf, but it is something that I really believe, which is that for any of these things you're doing, tracking the outcomes of that thing is very, very  important and will both be tactically helpful, but also I think, like you said, give you these inspirational moments that keep you going, whether that's awe or inspiration or fulfillment or whatever that feeling is that helps you keep going. I think that tracking the outputs of your work such that you can understand the impact that you have is both very strategic and the most rewarding way to do anything, I think”. Charna Parkey“Given the venture-backed nature of a lot of these startups, there's going to have to be some sort of monetization at some point. You're not gonna have 1 million, 10 million, 40 million dollars dumped into just giving software away for free. So sort of these misaligned motivations are certainly what raised my hackles where I'm like, oh, you're claiming forever or you're claiming that you're like a values-driven organization, but you're venture-backed and you need to make money. And so show me how those motivations align or misalign. Tell me what your monetization strategy is gonna be. I know you need one. That way I'm not wondering, should I use this? Should I not?” LinksConnect with CharnaConnect with Avi 

Timestamps00:00 - Intro05:10 - Paula’s Professional Journey10:30 - What Inspired Paula to Go Through the Open Source Path14:50 - What are some of the biggest challenges and impacts that Paula sees in companies trying to derive value?23:30 - Is the Tech World a Meritocracy? 25:35 - A Shift Of What is a Tech Company?27:30 - Kids Interacting with New Technologies31:30 - What Does Open Source Data Means to Paula? 42:50 - What is a Question that Paula has never been asked before?47:00 - What Advice would you give to the audience? 51:50 - Backstage with Executive Producer Leo Godoy Quotes:Charna Parkey“I think from my side, as the applications we build change, then some of those backing technologies have to. Where databases used to be used by expert-like database administrators and you needed to have like data architects to your data model and you had to do all of these very, very specific things. And now we have this Gen AI moment and all of a sudden all of these specialized vector databases, NoSQL databases, etc., need to be used by an average developer. So they just want an API and it has to work and it has to be fast. And so, over these different moments, different technologies came about or were evolved, but I think it might be the application that's actually driving the change instead of the technology itself opening”.Paula Paul“It still surprises people to hear that 90% of any given modern application is open source and then there's 10% custom code that, depending on your company, you own or not. And it just still amazes me that we have these open source projects like jQuery is a project of the OpenJS Foundation and it's in a tremendous amount of our ecommerce infrastructure. But it's a project that's maintained by a very small team of contributors. And, you know, if this were a commercial product, it would be like a $1,000,000,000 company. (...) The piece of work being done by the new foundation to help make sure that we have the healthy web and that it's secure is really important, because people, if I say Log4j, people that remember those days know how important it is to keep security vulnerabilities addressed.And that's a concern for me, that people don't pay more attention to this. I mean, if you had a commercial software product, you typically would pay 20% a year in maintenance fees. But as many of us know, sometimes you find a bug and you would just report the bug, but it might take years for that bug to get fixed in a commercial release.Whereas if it's open source, there are people out there who can jump on it. But it's really crazy that there's no funding for that or no public works through the government, given all the dependance and dependencies that we have on these open source assets.” LinksLinkedIn - Connect with CharnaLinkedin - Connect with Paula 

This episode features an interview with Charna Parkey, Real-Time AI Product and Strategy Leader at DataStax. Charna has been developing AI and ML products over the last 17 years and has worked with 90 of the Fortune 100 in her various roles. She is also a co-author and inventor on several patents.In this episode, Sam and Charna discuss handing over the role as host, Sam’s new startup journey, and how their thinking has evolved during the explosion of LLMs.-------------------“Now, it seems like we have this opportunity where the conversation and the place that society is at is different. Where we want to contribute to the right set of data when we talk open source data. We want to make sure that we have the right data to train this model in order to get the right outcome. We want to provide a lens of, ‘All right, you are this persona. How would you say this thing?’ I do think that from a lot of what the LLMs have today, the outcome of those words are still missing. And we need to solve that. Like, ‘Is this piece of writing actually going to achieve the outcome I want versus am I following legal's guidelines? Am I technically correct? Is my CEO going to like it?’ That doesn't mean you're achieving impact in the world. There's an aspect there where we've given feedback loops, it seems, to be like, ‘Did I like the answer or not?’ But not, ‘Did I take an action?’ As we get to autonomousness, we're going to have to have an outcome or multiple outcomes associated with the reward of the system.” – Charna Parkey“I personally believe that all cognition is bias. My degree is in cognitive science. One of the things that we trained on is attention. And to pay attention, literally means to selectively choose what data is coming in from the world that you're going to pay attention to and what you're going to discard. Which is also, to me, the definition of bias. All cognition is bias, but what do we care about? Do you trust this thing? What does that mean? Well, do you trust it to do these particular actions to a level of consistency in this particular domain? It doesn't mean that you're going to trust it in all environments. There's a lot more nuance that hopefully will evolve in this strange age of nuanced destruction machines.” – Sam Ramji-------------------Episode Timestamps:(01:04): Sam and Charna catch up (06:05): Sam explains his new company, Sailplane (14:21): How Charna’s thinking has evolved during the LLM explosion(25:45): Sam’s thoughts after 5 seasons of Open||Source||Data(38:52): What Charna is looking forward to in the next season of the podcast(40:44): A question Sam wishes to be asked(45:45): Backstage takeaways with executive producer, Audra Montenegro-------------------Links:LinkedIn - Connect with CharnaLinkedIn - Connect with SamLearn more about Sailplane

This episode features a panel discussion with Stefano Maffulli, Executive Director of the Open Source Initiative (OSI); and Stephen O’Grady, Co-founder of RedMonk. Stefano has decades of experience in open source advocacy. He co-founded the Italian chapter of Free Software Foundation Europe, built the developer community of the OpenStack Foundation, and led open source marketing teams at several international companies. Stephen has been an industry analyst for several decades and is author of the developer playbook, The New Kingmakers: How Developers Conquered the World.In this episode, Sam, Stefano, and Stephen discuss the intersection of open source and AI, good data for everyone, and open data foundations.-------------------“Internet Archive, Wikipedia, they have that mission to accumulate data. The OpenStreetMap is another big one with a lot of interesting data. It's a fascinating space, though. There are so many facets of the word ‘data.’ One of the reasons why open data is so hard to manage and hasn't had that same impact of open source is because, like Stephen, the stories that he was telling about the startups having a hard time assembling the mixing and matching, or modifying of data has a different connotation. It's completely different from being able to do the same with software.” – Stefano Maffulli“It's also not clear how said foundation would get buy-in. Because, as far as a lot of the model holders themselves, they've been able to do most of what they want already. What's the foundation really going to offer them? They've done what they wanted. Not having any inside information here, but just judging by the fact that they are willing to indemnify their users, they feel very confident legally in their stance. Therefore, it at least takes one of the major cards off the table for them.” – Stephen O’Grady-------------------Episode Timestamps:(01:44): What open source in the context of AI means to each guest(16:21): Stefano explains OSI’s opportunity to shine a light on models and teams(21:22): The next step of open source AI according to Stephen(25:38): Creating better definitions in order to modify software(33:09): The case of funding an open data foundation(42:31): The future of open source data(51:54): Executive producer, Audra Montenegro's backstage takeaways-------------------Links:LinkedIn - Connect with StefanoVisit Open Source InitiativeLinkedIn - Connect with StephenVisit RedMonk

This episode features a panel discussion with Mikiko Bazeley, Head of MLOps at Featureform; Zain Hasan, Senior Developer Advocate at Weaviate; and Tuana Celik, Developer Advocate at deepset.In this episode, Mikiko, Zain, and Tuana discuss what open source data means to them, how their companies fit into the AI-first ecosystem, and how jobs will need to evolve with the AI-native stack.-------------------“We're almost part of a fancy new AI robot kitchen that you'd find in Tokyo, in some ways. I see a virtual feature store as, yes, you can have a bunch of your ingredients tossed into a closet. Or, what you can do is you can essentially have a nice way to organize them. You can have a way to label them, to capture information.” – Mikiko Bazeley“I really like that analogy as well. I like how Mikiko put it where a vector search engine is really extracting value from what you've already got. [...] So where I see vector search engines, really, is if we think of these embedding providers as the translators to take all of our unstructured data and bring it into vector space into a common machine language, vector search engines are essentially the workhorses that allow us to compute and search over these objects in vectorized format. They're essentially the calculators of the AI stack.” – Zain Hasan“Haystack, I would really position as the kitchen. I need Mikiko to bring the apples. I need Zain to bring the pears. I need Hugging Face or OpenAI to bring the oranges to make a good fruit salad. But, Haystack will provide the spoons and the pans and the knives to make that into something that works together.” – Tuana Celik-------------------Episode Timestamps:(02:58): What open source data means to the panelists(09:11): What interested the panelists about AI/ML(24:10): Mikiko explains Featureform(27:00): Zain explains Weaviate(30:23): Tuana explains deepset(36:00): The panelists discuss how their companies fit into the AI-first ecosystem(44:58): How jobs need to evolve with the AI-native stack(54:35): Executive producer, Audra Montenegro's backstage takeaways-------------------Links:LinkedIn - Connect with MikikoVisit FeatureformLinkedIn - Connect with ZainVisit WeaviateLinkedIn - Connect with TuanaVisit deepsetVisit Data-centric AI

This episode features an interview with Mona Rakibe, CEO and Co-founder of Telmai, an AI-based data observability platform built for open architecture. Mona is a veteran in the data infrastructure space and has held engineering and product leadership positions that drove product innovation and growth strategies for startups and enterprises. She has served companies like Reltio, EMC, Oracle, and BEA where AI-driven solutions have played a pivotal role.In this episode, Sam sits down with Mona to discuss the application of LLMs, cleaning up data pipelines, and how we should think about data reliability.-------------------“When this push of large language model generative AI came in, the discussions shifted a little bit. People are more keen on, ‘How do I control the noise level in my data, in-stream, so that my model training is proper or is not very expensive, we have better precision?’ We had to shift a little bit that, ‘Can we separate this data in-stream for our users?’ Like good data, suspicious data, so they train it on little bit pre-processed data and they can optimize their costs. There's a lot that has changed from even people, their education level, but use cases also just within the last three years. Can we, as a tool, let users have some control and what they define as quality data reliability, and then monitor on those metrics was some of the things that we have done. That's how we think of data reliability. Full pipeline from ingestion to consumption, ability to have some human’s input in the system.” – Mona Rakibe-------------------Episode Timestamps:(01:04): The journey of Telmai (05:30): How we should think about data reliability, quality, and observability (13:37): What open source data means to Mona(15:34): How Mona guides people on cleaning up their data pipelines (26:08): LLMs in real life(30:37): A question Mona wishes to be asked(33:22): Mona’s advice for the audience(36:02): Backstage takeaways with executive producer, Audra Montenegro-------------------Links:LinkedIn - Connect with MonaLearn more about Telmai

This episode features an interview with Larry Augustin, angel investor and advisor to early-stage technology companies. Larry previously served as the Vice President for Applications at AWS, where he was responsible for application services like Pinpoint, Chime, and WorkSpaces.Before joining AWS, Larry was the CEO of SugarCRM, an open source CRM vendor. He also was the founder and CEO of VA Linux, where he launched SourceForge. Among the group who coined the term “open source”, Larry has sat on the boards of several open source and Linux organizations.In this episode, Sam and Larry discuss who owns the rights to data, the data in to data out ratio, and why Larry is an open source titan.-------------------"People are willing to give up so much of their personal information because they get an awful lot back. And privacy experts come along and say, ‘Well, you're taking all this personal information’. But then most people look at that and say, ‘But I get a lot of value back out of that.’ And it's this data ratio value question, which is: for a little in, I get a lot back. That becomes a key element in this. And I think there has to be some kind of similar thought process around open source data in general, which is if I contribute some data into this, I'm going to get a lot of value back. So this data in to data out ratio, I think it's an incredibly important one. And it gets everyone in the mindset of, ‘How do I provide more and more and take less and less?’ It's a principle of application development that I like a lot. And I think there's a similar concept here around open source data. Are there models or structures that we can come up with where people can contribute small amounts of data and as a result of that, they get back a lot of value.” – Larry Augustin-------------------Episode Timestamps:(02:52): How Larry is spending his time now after AWS(06:25): What drove Larry to open source(18:41): What is the GPL for data?(24:28): Areas of progress in open source data(28:57): The data in to data out ratio(36:39): Larry’s advice for folks in open source-------------------Links:LinkedIn - Connect with LarryTwitter - Follow Larry