Hongkai Wu

👉 Amazon OpenSearch Ingestion adds support for ingesting data from self-managed sources Amazon OpenSearch Ingestion now allows you to ingest data from self-managed OpenSearch, Elasticsearch and Apache Kafka clusters, eliminating the need to run and manage 3rd party tools like Logstash to migrate your data from self-managed sources into Amazon OpenSearch Service. Now you can seamlessly migrate or continuously replicate your data from all OpenSearch versions and Elasticsearch 7.x versions either on Amazon EC2 or on-premises environments into Amazon OpenSearch Service managed clusters or Serverless collections. #aws #amazonopensearch #seamlessingestion

6

Coffee & Systems (Study Group - Bengaluru) Week#5 (HPBN Speedread - Part 2 [HTTP]) December 22, 2024, 4:00 PM → 5:30 PM (Sunday) - The Market RMZ Ecoworld ------ Building on our discussion from last week, we'll explore the basic building blocks of HTTP. Since this is a comprehensive topic, we're splitting it across two weeks, with this session focusing on HTTP 1.x. Next week, we'll cover HTTP/2 and QUIC. - Brief History of HTTP - https://lnkd.in/gzPSvwwu - Primer on Web Performance - https://lnkd.in/gQnxzQkd - HTTP/1.x - https://hpbn.co/http1x/

37

1 Comment

Earlier this week, we announced an update to how Amazon Q Developer Software Agent works - https://lnkd.in/gVmTUQhA. Our efforts since making Q Developer Generally Available in May '24 have propelled the agent for software development to new heights. The SWE-bench scores have increased significantly, rising from 25.6% to 38.8% on the verified dataset and from 13.82% to 19.75% on the full dataset. This progress is a result of our customer-centric approach, incorporating developer feedback, and rapidly iterating to address pain points. What sets our agent apart is its architecture, which not only leverages the best LLMs but also equips the agent with the ability to explore multiple paths continuously to find optimal solutions for problems, similar to how developers work. There is a FREE version of Amazon Q Developer available (https://lnkd.in/gJdU3pB5) for you to try and experiment with. Install or update to the latest version of the Q Developer IDE extension, then type "/dev" in the chat window to put the Q Developer Agent to work.

37

bookmarking these cool series classes on LLMs in JAX.

73

Long-term memory is "just" a built-in document store that supports `get`, `put` and `search`. But that simplicity is for a reason. We see that memory is still application specific - and so want to give developers low-level building blocks to use In many ways, memory is similar to agents: - Both are highly hyped - Both are hard to define - For both, we see that application specific implementations (e.g. vertical) are winning

171

7 Comments

Exciting day ahead as the NVIDIA & Reliance Industries Limited AI Summit takes place today, bringing together two giants from the tech and business worlds! With Jensen Huang , CEO of NVIDIA, and Mukesh Ambani, Chairman of Reliance, leading the conversation, this summit is set to explore the cutting-edge innovations in AI and how they are shaping industries across the globe. From breakthroughs in AI computing to applications in 5G, cloud, and energy, the discussions here are going to shape the future of technology in India and beyond. Looking forward to insights on how AI will transform the digital landscape, accelerate innovation, and open up new opportunities for businesses everywhere. #NVIDIA #Reliance #AI #TechSummit #FutureOfAI #Innovation #AIRevolution #JensenHuang #MukeshAmbani #Summit

9

Check out the latest in our "Debuggability" blog series! This blog is authored by Kevin Downey Jason Johl and me and talks about how we built a Debuggability Paved Road at Intuit using Ephemeral Containers and Argo Workflows. We'd love to hear your thoughts on the topic. Also, what tools do you build as-a-platform to help debug service, compute and network issues in Kubernetes at scale? #intuit #kubernetes #debug

23

Recently Haytham Abuelfutuh presented about Batch Workloads and how Union and Flyte are used for Large scale batch and near realtime workloads. Check out this post to 1. Learn about the challenges in using Kubernetes for running Batch workloads 2. The problems in the programming model with Batch workloads. Often times its not a single job - but it may start with one 3. The simplification that can be introduced by using Union. Union runs them across multiple clouds, multiple kubernetes clusters, securely accessing large amounts of data, while maintaining efficiency through - caching, smart techniques in which container state is saved, how pods are used, how back pressure is maintained etc. Check out the blog.

9

How Uber Serves Over 40M Reads Per Second from Online Storage Using an Integrated Cache 1) Uber has in-house, distributed database built on top of MySQL, DocStore, which stores tens of PBs of data and serves tens of millions of requests/second. Utilized by microservices from all business verticals at Uber, Its use cases are growing, and so are the request volume. 2) So, for serving application with low latency and higher performance with scalability, Challenges like Disk Data Retrieval Speed Limit, vertical scaling, managing data partitions, request imbalances and increasing cost needs to be addressed. 3) Docstore has 2 layers a) stateless Query Engine layer - for query planning, routing, sharding, schema management, node health monitoring, request parsing, validation, and AuthN/AuthZ.  b) Storage Engine layer - for consensus via Raft, replication, transactions, concurrency control, and load management 4) Uber developed CacheFront, integrated caching solution for Docstore - redis on Query Engine layer. 5) Microservices can configure cache usage on a per-database, per-table, and even per-request basis to avoid cache invalidation. 6) Cached Reads: CacheFront uses a cache aside strategy to implement cached reads with TTL for cache invalidation. 7) Cache Write: Docstore has change data capture and streaming service "Flux". Flux tails MySQL binlog events for DB writes from storage engine layer to invalidates or upserts the new rows in Redis. 8) Timestamp is used as to serve versioning for avoiding Deduplicating Cache Writes Between Query Engine and Flux. 9) Having Dedicated APIs to explicitly invalidate the cached rows after the corresponding writes have completed for stronger consistency based on application needs. 10) To ensure warm caches across multiple regions, redis write streams are tailed to remote regions,which issues read requests to the query engine layer. 11) Non-existent rows are written to the cache with a special flag to ignore the row when querying the database and also don't return any data back to the user for the row. 12) Multiple Redis clusters are mapped to single Docstore instance to accomodate very large number of read-write requests.In case of a redis cluster failure,all requests from a failed Redis shard will be distributed among all database shards. Reference - https://lnkd.in/g_iQHTSA Feel free to add any relevant information in comments section #systemdesign #uber #redis #cache #docstore

22

2 Comments

Have been brewing up abstractions on Langgraph(from LangChain)! What if there existed a nifty SDK, that allowed you to invoke any agentic design pattern? Why is it that I need to understand "Reflection", why is it that I need to understand "Hierarchical agents"? Possible to improve dev experience, improve the robustness of these agents, and not compromise on configurability? Spun up a quick Loom to braindump this, anyone thinking along these lines?

24

7 Comments

Distributed transactions can be hard. Reading this beautiful paper by Dynamo Db team https://lnkd.in/g8rXxVzT to learn how probably the world's biggest and busiest database handles this.

2

A good use case for using context-caching, specially if you are trying to build RAG on a large cluster of documents. https://lnkd.in/gTEaG_ET

3

VPC with zone subnets, alb, nat and igw https://lnkd.in/gZxkgdBx

5

The https://microservices.io/ patterns cover different areas of microservice architecture: service decomposition, data consistency, service communication, deployment, and reliability. I recommend the following: Decompose by Business Capability and Subdomain: Define each service around specific business capabilities and use DDD for clear boundaries. Database Per Service: Ensure each microservice has an isolated data store. Saga Pattern for Consistency: Use sagas for handling distributed transactions, employing both choreography (event-driven pub-sub) and orchestration as needed. This is the most interesting for me for now! API Gateway and Backend for Frontend (BFF): Manage client access and tailor specific frontends. Circuit Breakers and Health Checks: Maintain reliability and resilience. #grekai #microservices #architecture

5

Successfully fine-tuned llama3.2 3b on custom data, generating domain-specific QA data from public sources. Integrated "Wh," "Yes/No," and "Reasoning" question types for top-notch performance. All done on Google Colab and inference using a general-purpose CPU! #AI #MachineLearning #DataScience #TechInnovation #GenerativeAI #Finetuning

21

1 Comment

At the heart of PreProd OpnSrc, there is a vision and a mission to make foundational learning at ZERO cost Please visit the post below and the page at PreProd Opnsrc for more information PreProd Corp PreProd Emulet PreProd Consult Krishnav Dave

10

Hey folks, diving into Kubeflow Pipelines with Google Cloud VertexAI? Let's talk architecture! Whether you're into the open-source vibe or prefer native GCP services, there's a pattern for you. Check out the supported GCP integrations and let's elevate those workflows together!

15

1 Comment

If you are using AI Agents, here is a curated list of top frameworks to consider. (Note that the order is irrelevant). - LangGraph (https://lnkd.in/g7_wCcwM) - CrewAI (https://lnkd.in/g8Vk832N) - AutoGen (https://lnkd.in/gbVQVaay) - Pydantic AI (https://lnkd.in/gQj7Qave) - Smol Agents (https://lnkd.in/gwWYq7Ji) - Phidata (https://lnkd.in/gJDi3seG) #AI #GenerativeAI #Agents

27

The Kafka Connect file-chunk connectors (now available on Confluent Hub) can be used to stream files (of any size) through a Kafka topic. Stream audio, video, images, documents - content of any type. This article explains why (and when) file-streaming may be needed; and the original purpose of the connectors.

27