anais9

Hearst | Senior Engineer, Generative AI | New York, NYC | Hybrid/Remote

Here at Hearst we've been building out generative AI use cases and fluency across our portfolio of businesses spanning well known media brands to healthcare, transportation, and financial services. While we are an established corporation with multi-national presence, the mission and support behind our generative AI team more closely resembles a well-funded early stage startup as we engage in radical business transformation empowered by nascent technology.

The ideal candidate is one who has written code but now builds systems largely leveraging agentic development tools such as Claude Code (w/ Opus 4.5) & Codex (w/ GPT-5.2), and who sees the future of development as managing swarms of AI agents rather than cranking out code by hand.

Come join our team and help flesh out the future of business on a global sale while getting deep hands-on experience with the latest and greatest tech!

Apply via https://eevd.fa.us6.oraclecloud.com/hcmUI/CandidateExperienc... or feel free to reach out to me directly at alex dot redmon at hearst dot com.

Hearst | New York City - Hybrid or Remote | Full-time | Senior Engineer, Generative AI & Senior Program Manager, Generative AI

Hearst's corporate team is looking to fill two roles (in Engineering & Program Management) on its Generative AI Team, working to empower businesses across our diversified portfolio spanning publishing, health, transportation, and financial services. Our team's purview includes evaluating and benchmarking new releases, experimenting with the best ways to apply them to existing business problems, piloting POCs that demonstrate new ways of doing things, and accelerating the entire process on our team and within our various businesses through AI driven development tools & tactics.

Do you enjoy actively learning and experimenting with new tools & processes? Do you like solving problems across a wide range of domains and disciplines? Are you excited by the potential and application of emergent generative capabilities and agentic autonomous systems? Check out one of our open postings or feel free to reach out to me directly (https://www.linkedin.com/in/alexandria-redmon-5078865/):

Sr Engineer: https://eevd.fa.us6.oraclecloud.com/hcmUI/CandidateExperienc...

Sr Program Manager: https://eevd.fa.us6.oraclecloud.com/hcmUI/CandidateExperienc...

I envy your rig - mine glitched a lot to get it in <3min. Might not be doing myself a service by actually answering the Duolingo questions via LLM... https://www.youtube.com/watch?v=I-J0ppP-H9s

Wanted to see how quickly I could get to the beach - was limited by my system resources and an inability to load the closing video. Auto-solving the Duolingo questions was clearly the best part.

I love it! Here's my stab at an automated speedrun: https://www.youtube.com/watch?v=4z8P1F-lyB8

A progress meter may not be necessary, but as the article points out if you have comments or an otherwise large footer not associated with the content of the post, the scrollbar can be deceiving.

I'm personally a huge fan of the progress meter (having once thought it was redundant as well) - one other easy addition I didn't see mentioned is an "estimated reading time." Having a ballpark range for how long I should expect to spend with a piece of content greatly increases my chance of engaging with it, and the progress meter creates a tangible representation of that time (and how much of it I have left to finish consuming the content).

This is awesome! Until GPT-4o dropped, Claude 3 Opus was hands down my go-to for code generation.

Between these model performance improvements and their new "artifacts" handling, I get the impression this update may sway me strongly back towards Anthropic (at least for this use case).

Anyone here used this yet? I've been a big fan of their Gen-2 offering so excited to get my hands on longer generations with higher quality and consistency

Started working on something to build complete multi-file changes via LLM and chose to write it as a CLI tool both for convenience and ability to compose feature builds via scripts and other tools.

It supports various providers (OpenAI, Anthropic, Azure, Gemini), supplying both local and remote resources (defined by arguments and/or configuration file directives), and allows for feedback after initial implementation which is useful for troubleshooting issues in initial tack.

A few things I plan on adding soon include: - Improved source mapping - Hands-free voice mode - Image processing

Would love to see Ollama support for this - seems promising given my experience with LLaVA so far and would love to get some hands on head to head experience

While your point about Docker’s primary purpose is valid, containerization is commonly used for security isolation as well. With proper configuration, it can be very useful towards this end.

Can you suggest any preferred alternative methods of isolation that offer similar efficacy and ease of use for quickly running complete software systems made by an unknown/untrusted actor?

It goes down in the DMs https://x.com/tim_cook

CrayEye is an open-source (Flutter/Dart) multimodal LLM visual analysis utility. made to build & share AI vision prompts augmented with native device sensor data.

Homepage: https://www.crayeye.com

GitHub: https://github.com/alexdredmon/crayeye

iOS: https://apps.apple.com/us/app/crayeye/id6480090992

Android: https://play.google.com/store/apps/details?id=com.crayeye.ap...

This app was made with AI - read more about that via https://www.alexandriaredmon.com/blog/the-app-that-ai-made

I (or more accurately A.I.) made the FOSS mobile app https://www.crayeye.com to make it easier to experiment with multimodal vision prompts augmented by device data (e.g. location, date/time).

While this tool still uses GPT-4v / GPT-4o as its default, it now supports configuring custom engines (via OpenAPI spec) which can point to any API/model - this has been tested using Llava (and Bakllava) running locally via Ollama.

Sharing an open-source app built by A.I. and the process of making it.

Resources:

- Project Homepage: https://www.crayeye.com

- GitHub Repo: https://github.com/alexdredmon/crayeye

- iOS: https://apps.apple.com/us/app/crayeye/id6480090992

- Android: https://play.google.com/store/apps/details?id=com.crayeye.ap...

- Blog post: https://www.alexandriaredmon.com/blog/the-app-that-ai-made

Yeah, my interest was piqued by the title but that demo video torpedoed it... Which is such a bummer because language learning is a perfect use case for LLMs

Completely agree - I made https://notchbegone.com to help mask it visually when it came out expecting Apple to quickly obviate via their own design updates and yet years later we're still having this conversation.

My tinfoil hat take is that the true primary purpose of the notch is to drive upgrade purchases at some point in the future whenever serious technical advancements are waning.

Totally - for what it's worth, if you ask it for the Python source code that it's running to create the .GIF it will provide it.

Also I'm glad someone else appreciates the pun :)

The prompt used:

You are not a personal assistant, you do not answer questions. You have one purpose and one purpose only - follow the following instructions:

- Whatever the user says to you, consider this your "scene." Your objective will be to generate an animated .GIF of this scene.

- First, break the scene down into a distinct beginning, middle, and end. After doing this, consider what the transitions between beginning, middle, and end for the scene are. List visual descriptions for each of these frames, taking careful consideration to keep everything in the frame consistent by explicitly describing everything that should remain the same in great detail across each frame except for what is moving or changing - i.e. beginning, transition 1, middle, transition 2, end

- Create DALL-E images for each of these frames

- Write a Python script to combine these images into a single animated .GIF file and present the user with a link to download their image

- Ask the user for their next scene prompt and then begin the process over again from the beginning

A simple interactive ChatGPT version of one of my favorite NPR radio shows "Wait Wait... Don't Tell Me!"

Uses the AP News feed of oddities to play "Bluff the Chatter" and the "Chatter Limerick Challenge"

Have been working with this and very impressed so far - it’s a step ahead of LangChain agents and seems to be receiving more attention/development than LangChain was interested in committing to agents.

FWIW the “group research” and “chess” examples from the notebooks folder in their repo have been the best for explaining the utility of this tech to others - the meme generator does a good job showing functions stripped down but misses a lot of the important bits

Thank you for expanding your experience, tackling personal growth, and putting it into words for others to share.

Something that gets overlooked so frequently in this conversation is that the very traits of a patriarchal system that are purportedly positive for the organization they're instilled in are inherently crippling.

Being overbalanced on "strong," "commanding," and even "risk taking" won't help you innovate in the long run if you actively eschew collaboration and empathy and the individuals whom exemplify the traits you lack.

Corresponding Google Colab notebook: https://colab.research.google.com/github/githubpradeep/noteb...

Autogen repo: https://github.com/microsoft/autogen

I created this Twilio flow months ago to converse with ChatGPT audibly (and text message it) and wanted to share with the community.

Requirements:

- Twilio account

- Hosting for API / serverless function

I need to check out Tab Stash - while I'd played around with using TabFS for this use case, it was a little bit clunky. Totally agree with your point about global sort/unique which is the basis for my current bookmarking system of https://save.page/docs/saving

For sure, but even a lot of those sites maintain a favicon.ico to support legacy crawlers/systems - then for those that don't (i.e. trigger onError), falling back to DDG's icon is an easy way to skip all that parsing.

This - an alternative approach could be to create the images with a URL of '{domain}/favicon.ico' and then add an onError event to switch the URL to the DDG URL only for instances where the site's favicon isn't in the default location.

I wrote this because like others, I found the "notch" on newer iPhones and MacBooks distracting - particularly when using a second display as a primary monitor (when there's no menu bar to balance out the unusable space at the top of the screen).

I realized that by creating a tinted bar the same height as the notch across the top of a background image, I could significantly diminish the distraction factor of the notch. Someone had written an app that did something like this, but I wasn't interested in keeping some piece of third party software running all the time just to achieve this.

Enter: notchbegone.com. By writing a tool to auto-generate background images with my desired top bar overlay, I could easily subdue the notch without any added overhead to my system.

Tool: https://notchbegone.com

Source: https://github.com/alexdredmon/notchbegone

I got addicted to del.icio.us in its heyday, so once it got acquired and subsequently sunset I wrote https://savecrate.com and haven't looked back. The ability to save anything (public/private) without having to weigh whether it's "important enough" and retrieve it quickly via search/tags (eliminating the necessity of messy folder/hierarchical structures) is indispensable to my workflow.

Cylera | Full Stack Software Engineer | New York City (Flatiron) | ONSITE | Full-Time | https://www.cylera.com/

Cylera is a VC-backed startup creating a network-level cybersecurity solution to manage and protect connected medical devices, such as infusion pumps and CT machines. Our problem space is at the intersection of the virtual and physical, where insecure medical equipment serves as the interface between patients and the hospital network, and where a single device compromise could have potentially life-threatening consequences.

Our stack largely consists of Python (Flask API and pipeline/analysis services) and React (customer and administrative dashboards) running on GKE managed by Helm & Terraform, with data stored in systems including Postgres, BigQuery, and Elasticsearch to name a few.

We're looking for someone who has:

- Proficiency in both Python and JavaScript

- Experience with a Single Page Application framework, preferably React

- Experience with a modern web server framework, preferably Flask

- Professional experience using relational databases, preferably Postgres

- Experience with Docker and container systems

- Experience with unit/integration testing and CI/CD pipelines

- Experience using a cloud platform, preferably Google Cloud Platform

Apply via https://boards.greenhouse.io/cylera/jobs/4450659002 or view all of our open postings at https://www.cylera.com/careers-application