for now…🤓

The more I play around with GEPA the more I realize its not just a prompt optimization algorithm. Its really the most efficient way to have an LLM explore a massive dataset and and make useful insights. This visualizer alone shows just how cool that process is.

Open source btw! github.com/modaic-ai/modaic/…

follow ML best practices in your optimization runs with the modaic sdk modaic.dev/blog/modaic-bring…

Next episode in the DSPy Series will be with two young entrepreneurs @FaroukAdeleke3 and @ty_todd1 building @modaicdev which makes distributing your DSPy code easy. Subscribe youtube.com/@Informationshel… to get the updates

Holistic Benchmarking with dspy-helm from Asad Aali on @modaicdev ! ☺️ Paper: arxiv.org/pdf/2511.20836 Github: github.com/StanfordMIMI/dspy… Modaic: modaic.dev/asadaali

Inspired by the DSPy Codex, I built a @DSPyOSS scaffold for Claude Code on @modaicdev ! It shares full feature parity with Claude Agent SDK, including MCP servers, tool use, hooks, etc. but with a DSPy signature interface. Links below. x.com/AlexGDimakis/status/19…

Modaic and Weaviate 1. Modaic and Weaviate: Load the `CrossEncoderRanker` program from the Modaic Hub, as well as `PromptToSignature` (github.com/weaviate/recipes/…)

December is the month for giving So here are 9 new Recipes demoing how to use @weaviate_io and @awscloud, @ContextualAI, @ChonkieAI, @p0, @Google, @modaicdev, @twelve_labs, and @llama_index 🧑‍🍳

About a year ago, @plasticlabs achieved SOTA with DSPy on the OpenToM benchmark. The benchmark tests models’ ability to track and reason about the beliefs, perceptions, intentions, and psychological states of simulated characters (social cognition). @vintrotweets experiments are now packaged and available on @modaicdev with the ability to run one of his optimized programs with your own variables and under your own evaluations. Links below.

hmmm maybe like a package registry for DSPy programs…wonder if someone’s working on this hmmm

Most leaderboards use one fixed zero-shot prompt across all models. Problem: Different LMs have different “prompt ceilings.” When you give them all the same prompt, it becomes less about benchmarking models and more about benchmarking how that model performs on a fixed prompt. Solution: Stanford shows that adding structured prompting (especially zero-shot CoT via @DSPyOSS) lifts performance by ~4 points on average and can even flip model rankings on tasks like MMLU-Pro, GSM8K, and MedCalc.

Extremely underrated paper out of Stanford, including one of the creators of MEDVal! Benchmarking the ceiling of LM's as systems (optimized prompting strategy LM) instead of generalizing the same prompt over competing LM's proves to be a more holistic evaluation of their capabilities. arxiv.org/pdf/2511.20836

game changer.

Hot take: retrofitting RL into your agent is like taping a jet engine onto a bicycle. An RL-native agent development kit changes the entire build process-cleaner code, faster iteration, agents that are actually self-improving. The difference? Night and day.

new blog! - software finally eats labor - ai applications as an operating system - we’re in a fuzzy state in ai engineering - necessary abstractions drive innovation - why an open source ecosystem needs to exist at the right abstractions modaic.dev/blog/prompts-are-…

This is an agentic people search I built with @DSPyOSS and @modaicdev. It works by having a user input their company description and their ICP (Ideal Customer Profile). The LLM then generates an initial search for Persana and returns some profiles. The user can refine the search by selecting the profiles they like and giving some feedback. This continues for multiple iterations until the user is satisfied with the search results. Offline, the set of selected profiles acts as a reward function. (how many of the target profiles did the generated search parameters select). The content of the unselected target profiles provides valuable feedback about what the search parameters missed. The reward function and this feedback is fed into (you guessed it) GEPA, to optimize the prompt to generate better search parameters the first time around. swagginty/persana-lead-gen is public on Modaic Hub!! modaic.dev/swagginty/persana…