Ankur Goyal

I’m excited to share a new benchmark, MIPfeas, which shifts the focus in Mixed-Integer Programming from "final optimality" to "early feasibility." Using the Primal Integral metric, we evaluated how solvers like HiGHS, SCIP, and NVIDIA cuOpt deliver high-quality feasible solutions within a short time window. While direct comparison with commercial MIP codes remains difficult, we have grouped the results of three major commercial solvers—COPT, CPLEX, and Xpress—into three synthetic baselines: the Virtual Best, Mean, and Worst Commercial Solvers in this benchmark. A huge thank you to Hans Mittelmann, the NVIDIA team (Burcin Bozkaya, Akif Çördük, and Chris Maes), and ZIB Zuse Institute Berlin for the technical collaboration. Special thanks also to Timo Berthold, Gerald Gamrath, and Ferenc Katai for their expert insights. Check out the full results here: https://lnkd.in/dSfNc58v #Optimization #MIP #GAMS

83

11 Comments

AI-assisted coding isn’t about fewer keystrokes, but greater clarity, deeper thought, and scalable precision. Diwank Singh’s "Field Notes From Shipping Real Code With Claude" guides developers through battle-tested AI workflows: 🍃 Playground Mode for swift exploration. 📐 Pair Programming Mode using CLAUDE.md to shape architecture and uphold style. 🚦 Production Mode fortified with human judgment, stringent tests, and anchor comments. This is developers orchestrating AI to amplify craftsmanship. If you're exploring how AI can elevate your coding practice, start here. Read more → https://lnkd.in/gG68gjxr #AICode #EngineeringBestPractices #DevOps

8

3 Comments

YellowDog.ai just set a 10x benchmark uplift in scale computing, delivering 40,000 tasks per second (TPS) and managing 100,000 compute nodes in the cloud. What's even more interesting, that's 2x IBM Symphony and opens an intriguing pathway for these until-know closed/captive systems.

11

Building a production-ready streaming user embedding pipeline in ~5 minutes with Zipline + Claude code: https://lnkd.in/gnWsjrbn Why is Zipline + Claude Code such an effective combo? 1) Chronon offers a simple API that unifies batch + streaming computation, online serving and offline backfilling, and allows for flexible aggregations, transformations and embeddings. It automates the complex infrastructure involved in building these pipelines, allowing the AI to focus on getting the semantics right (which is what it’s good at). 2) The Zipline skill gives the AI a few critical tools: the ability to read metadata about various data sources (giving it starting schemas to build pipelines), plus the ability to call `eval` on the Zipline entities that it creates, which allows it to verify that everything that it creates is indeed correct quickly without needing to any data processing. The result is that you build powerful pipelines easier than ever before. Something that would have taken a team of engineers weeks to do can now be done with a single prompt in 5 minutes. Check it out in this brief demo.

47

First piece on the "AI Native Dev". Read up on how we make Claude (and other agents) better at Go with Tessl's tech!

19

We shipped DeepEval + LangGraph integration. Two lines of code to run eval metrics on your LangGraph agents. People kept telling us LLM-as-a-judge setups require too much boilerplate. Prompt templates, parsing logic, retry handling, score normalization. By the time you're done wiring it together, you've built half a framework. It should be simple. It should be reliable. That's the point of DeepEval. Now with LangGraph: trace your agent, apply metrics, see what breaks. No ceremony. Link below 👇

33

7 Comments

Juan Sequeda started a good thread on #ContextGraphs, which should lead to an interesting conversation. Here's my comment - As to Context Graphs, the Foundation Capital post was very narrow in its scope re: capturing informal decisions from communication and collaboration software (largely unstructured text). Of course, it's just accounting for yet another source of operational information that can support audits, and if ontologically mapped, expands scope of reasoning, learning and adapting. #AI is raising expectations for contextualization and continuous optimization, which is implicitly driving demand for a unified model. Without a common #abstraction across the growing number of incorporated sources this is going to be a discombobulated disaster, alongside the emerging AgenticAI memory models. The problem is that while #GenAI and #AgenticAI co-opt Computer Science and Distributed Systems terms re: #Reflection, #Recursion, #SharedMemory, etc. - they implement them in the most trivial and naive way. The result is that the #DeepLearning crowd, in their drive for "Data uber Alles" is undermining the reasoning, planning and expressive power of the resulting solutions. We need to push back hard to in 2026 to make sure that the Data Scientists "gone wild" don't steam roll us with a foundational, but errant approach.

7

Lambdas are a low friction but expensive way to consume Kinesis. Not only is Lambda compute > 8x more expensive than equivalent EC2 under continuous use, but you pay per invocation. To consume a 1 million records / s stream with a Lambda, you're looking at 2000 invocations per second, each doing a batch of 500 records per invocation and taking (say) 100ms. This costs ~ $44k per month. There's huge savings just by consuming with regular EC2 compute. However, a lot of Kinesis use is as an intermediary channel, with data ultimately going to S3. Storing streams directly in S3 can be both cheaper and operationally simpler. That's what we do here, while keeping the same API for producers and consumers: https://lnkd.in/gA3raSJa

21

3 Comments

Last quarter, we shipped query tracing. Every resolver and model call is now instrumented and timed. The trace shows you exactly where time is spent in your query execution - which resolvers ran, what they called, how long each piece took. You can enable it per query with --trace in the CLI, or configure sampling to run automatically in production. For high-volume systems, sampling gives you continuous visibility without adding overhead. When you have a tight latency budget, you need to know where every millisecond goes. Read more in our product update in the link below.

34

1 Comment

I made something for all the .NET devs out there: a C# SDK for various AI providers. The GitHub repo is at https://lnkd.in/g2ESrazb and the Nuget package can be found at https://lnkd.in/gjnJW8Ts. This is something that's been on the to do list for awhile now, as much of this code helped build a good portion of the core of Zatomic AI. I knew pretty early that I'd eventually extract this out and make it available to everyone else, and I finally set aside some time to do that. The following AI providers are supported: AI21 Labs, Amazon Bedrock, Anthropic, Azure OpenAI, Azure Serverless, Cohere, Deep Infra Inc., Fireworks AI, Google Gemini, Hugging Face, Hyperbolic, Lambda, Meta, Mistral AI, OpenAI, Together AI, and xAI. The library calls the chat completions REST APIs and inference endpoints for each of the above AI providers. Everything is strongly-typed with the library handling all JSON serialization/deserialization for all requests and responses. And both non-stream and streaming functionality is supported. The repo includes a samples project that you can use to get going pretty quick. Much more in terms of documentation needs to be done, but this at least gets the ball rolling. If you know someone using .NET for AI development, please pass this along. And of course, let me know if you have any feedback.

27

3 Comments

AI-tools are everywhere, but they require thoughtfulness, creativity, and grit to use well. Josh continues to blaze the trail on this at Encamp. If you're interested in getting the most of AI development in an existing and fast-moving code base, I highly recommend reading this post.

9

Great piece by Emilia David on the LangChain open source integration ecosystem, the power of LangGraph, and our newly released LangGraph Platform Helps put everything you see us working in perspective Link: https://lnkd.in/ghajS7zx

100

5 Comments