Sandy (@sandykumar93) header
Sandy (@sandykumar93) profile picture

Developer, helping the world in building serverless applications, Principal Solutions Architect @AntStack

bangalore, india
Joined January 2011
Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)Photo by Sandy (@sandykumar93)
Ready to see AI in action? Join @sandykumar93 at #ServerlessDaysBLR2024 as he simplifies #RAG and #AmazonBedrock for real-world use! 🎟️ Last-minute tickets available, grab yours now 👉 konfhub.com/sdb 👈 #FridayTech #TechConferences #BangaloreEvents
3
8
275
Join @sandykumar93 at #ServerlessDaysBLR2024 to simplify LLMs, Prompt Engineering, and RAG with Amazon Bedrock. Learn to integrate AI seamlessly into your apps. 🗓August 31st, 2024 🎟️Buy now: konfhub.com/sdb #TuesdayTech #TechConferences #BangaloreEvents
1
2
181
and for coding!
64
179
108
1,678
561,353
Want to ship and deploy faster? Web agencies, developers, plugin creators—this is for you Join us at #DevStudioCon2024🇮🇳 and come see how you can 🤝 Take on more clients with less manpower 💰 Earn more with apps and widgets 🪩 Network and have fun Save your spot now

1
9
302
1,041,244
Gen-AI is not for me! It sucks! linkedin.com/pulse/gen-ai-me…
1
6
165
I hear a lot of criticism of the work I did on DynamoDB Single Table Design these days. I even saw @alexbdebrie apologizing for being an advocate the other day. The core concept behind STD, however, has always been the same as it is for all #NoSQL databases. What is accessed together should be stored together, and how that data is stored should be influenced by how it is accessed. Anyone who tells you a different story than the one that follows is wrong. I led the team that invented the Single Table Design pattern. I have facts, they have at best half-informed opinions. There are 3 people in this world who could tell this story. The other two I am sure could add more color, but the picture they draw will look the same. 8 years ago we had 5 GSI's, every index had its own capacity allocation, and there was no such thing as on-demand pricing. Even when GSI's were increased to 25 you still had to allocate capacity for each one. It was a major pain. Eventually all of these things were fixed, but to make things work before any of that happened we introduced a design pattern called Index Overloading to our data modeling guidance on the NoSQL Blackbelt team at Amazon. Summarizing the pattern, an index only needs to know what attributes to use as alternate partition and sort keys. If users specify generic attribute names like "GSI1PK" and "GSI1SK" then different Item types could store different values in those attributes and be indexed using type specific sorting on shared GSI's. The concept was a workaround that enabled teams to index each type of Item stored on the table 5 different ways, effectively eliminating the 5 GSI limit. The drawback of doing this was that indexes became more and more polluted with unrelated Item types as the number of access patterns they supported increased. Because of this it was not easy to drop and recreate indexes without table scans and batch updates which became expensive at scale. The pattern also introduced heavy cognitive load on developers as using abstract naming for index attributes meant it was not always immediately apparent when looking at the data how the data was being indexed unless the values assigned to the generic keys were self-explanatory. All of these things were tradeoffs for applying the Index Overloading pattern, not core issues with Single Table Design itself. They were often deemed acceptable inconveniences considering the benefit of having effectively unlimited GSI's. Most of the problems that drove the need for Index Overloading have been resolved over the years as DynamoDB has added support for 25 GSI's, introduced on demand pricing, and eliminated the need to allocate capacity individually for each index. As a result the pattern should really be considered deprecated today. Additionally, many people over the years have also taken STD to an extreme that was never intended. Mixing configuration and operational data, maintaining a single table across service boundaries, or storing unrelated data that is not accessed together in the same table. Despite the fact that there are some people out there trying very hard to rewrite history around this, none of these things were ever recommended as best practices. It is easy to look at the product as it exists today and criticize the design patterns of yesterday that were invented to deal with API deficiencies that no longer exist. I read some serious garbage every now and then written by people I feel should know better, Those people really had very little exposure to the process of solving the problems we faced when the patterns they criticize were introduced. Would I tell teams today to do some of the things I advocated for 6 years ago? Of course not. Does that make guidance we gave invalid years before features were introduced obviating that guidance? Of course not, and none of that invalidates Single Table Design. The core concepts of STD cannot be ignored without incurring cost and performance hits. It is the same with all #NoSQL databases: Data that is accessed together should be stored together, either embedded in the same Item or stored on the same table/collection where it can be easily indexed. To maximize efficiency of the system, the data structures used should be influenced by how the application accesses the data.
16
33
15
306
92,974
Introducing 'Prompt Engineering with Llama 2' — an interactive guide covering prompt engineering & best practices for developers, researchers & enthusiasts working with large language models. Access the notebook in the llama-recipes repo ➡️ bit.ly/3Hu266D
31
452
42
2,004
236,441
Hard to believe AWS Lambda was announced 9 years ago at #reinvent 2014. One of the most impactful cloud services (and communities) ever created. #serverless
3
12
4
84
19,490
Olympus is Amazon's new 2 trillion parameter LLM. IMO, super large language models are unwieldy, super inefficient to run inference on, and unnecessarily expensive The era of super-large LLMs is likely ending, and Amazon is just a little too late to the party. Instead, training on large, high-quality datasets and building models in the ~100B parameter range makes sense. Rumors are already floating around that GPT-4 is not one super-huge LLM but a mixture of experts (MoE) ensemble of several smaller models. If I were Amazon, I would take a page from Meta and train a multi-modal Llama-3 equivalent in the 100B range. That way, they can match GPT-4's performance at a fraction of the cost. Olympus will likely be announced at re:Invent later this month and will contribute to further commoditizing LLMs.
37
56
10
407
144,063
Sandy retweeted
We're rolling out new features and improvements that developers have been asking for: 1. Our new model GPT-4 Turbo supports 128K context and has fresher knowledge than GPT-4. Its input and output tokens are respectively 3× and 2× less expensive than GPT-4. It’s available now to all developers in preview. 2. Assistants API and new tools (Retrieval, Code Interpreter) will help developers build world-class AI assistants within their own apps. 3. The platform is becoming multimodal. GPT-4 Turbo with Vision, DALL·E 3, and text-to-speech are all now available to developers. Oh… and we’re doubling GPT-4 rate limits. openai.com/blog/new-models-a…
863
2,674
782
14,280
3,967,312
I have been threatening to write this for ages. Contrary to Betteridge's law, the answer is YES, most of the time 😅 We examine arguments for and against a monolith Lambda API function (aka Lambdalith) rehanvdm.com/blog/should-you…
6
4
2
34
7,484
Next talk is by our own fellow ant @sandykumar93 - Supercharge User Experience using Lambda Response Streaming! #ServerlessDays #ServerlessDaysBlr
1
9
277
It was awesome catching up with @davidand393! Always pleasure talking to people who talk same language!
6
24
1,887
🌟Silver Sponsorship Spotlight🌟 We are absolutely thrilled to express our heartfelt gratitude to @googledevs @googledevgroups for their incredible and wonderful support as a Silver Sponsor for #ServerlessDaysAhmedabad 2023! #ServerlessAhmedabad #ServerlessDays #Ahmedabad
3
3
212
I am speaking at reInvent this year! My first session API302 is an updated version of "Building Next Gen Applications with EDA". If you're looking for a fun and informative 300 level on Event Driven Architecture, I'd love to see you there. Be sure to star it so we get a big room!
3
3
25
2,484
📢Announcement! ServerlessDays Bengaluru Venue & our new supporter 🙌 We are thrilled to announce that ServerlessDays Bengaluru 2023 is wholeheartedly supported by @saplabsindia, fostering a vibrant serverless community. 🎟️ Register Now! lnkd.in/gyeAbVFx #Serverless
1
4
97
🥁 Exciting News 🥁 We are absolutely thrilled to unveil the Gold Sponsors who are supporting this year's Serverless Days Ahmedabad 2023! ✨@AntStack✨ 📅 Date: 9th September 📅 🎟️ Grab your Tickets : buff.ly/3pqcmHy #ServerlessDaysAhmedabad #Serverless #Ahmedabad
2
9
308