Synthropic | Detection Engineer + GenAI | Full-time / Part-time | Remote (SF)
We're an early-stage startup looking for someone passionate about detection engineering and generative AI.
If you know SIEM, threat intel, logs, detection modeling/strategy, purple teaming, and adversary emulation — and you've been wanting to explore how LLMs/agents can accelerate detection engineering workflows — we'd love to talk.
The role A lot of it is what we call AI babysitting: guiding 1,000 kindergarteners (LLMs) as they try to become senior detection engineers. Sometimes they're brilliant, sometimes chaotic — your job is to shape that into something production-ready.
You'll be part detection engineer, part researcher, part builder, with a big say in shaping the product and how AI gets applied to real-world problems.
Details Full-time or part-time Remote-friendly (SF-based, local a plus but not required) Early team, high ownership, fast iteration
We're an early-stage startup looking for someone passionate about detection engineering and generative AI.
If you know SIEM, threat intel, logs, detection modeling/strategy, purple teaming, and adversary emulation — and you've been wanting to explore how LLMs/agents can accelerate detection engineering workflows — we'd love to talk.
The role
A lot of it is what we call AI babysitting: guiding 1,000 kindergarteners (LLMs) as they try to become senior detection engineers. Sometimes they're brilliant, sometimes chaotic — your job is to shape that into something production-ready.
You'll be part detection engineer, part researcher, part builder, with a big say in shaping the product and how AI gets applied to real-world problems.
Details
Full-time or part-time
Remote-friendly (SF-based, local a plus but not required)
Early team, high ownership, fast iteration
Yeah MCP isn't really doing a whole lot. You can give an LLM a generic HTTP extension. Then list a series of GET/POST/PUT and ask it to form the calls and parse the response. The problem is its not really ideal as the calls aren't natural language and its common for it to misguess the next token and mess up things like the route, body, headers, with a hallucination. So people started shortening these calls to simple things like read_file, etc. Prior to MCP there was a ton of playgrounds doing this with simple Playwright functions.
The thing that surprises me is with MCP we have shirked all of the existing tools around OpenAPI specs, OIDC, etc. We could have created a system where all 'services' expose a mcp.slack.com/definition endpoint or something that spit back a list of shortcut terms like send_message and a translation function that composes it into the correct HTTP API (what most MCP servers do). For security we could have had the LLM establish its identity via all our existing systems like OIDC that combine authentication and authorization.
In the system above you would not "install an mcp package" as in a code repo or server. Instead you would allow your LLM to access slack, it would then prompt you to login via OIDC and establish your identity and access level. Then it would grab the OpenAPI spec (machine readable) and the LLM focused shortcuts 'send_message', 'read_message', etc. LLM composes 'send_message Hello World' -> translates to HTTP POST slack.com/message or whatever and bob's your uncle.
If you wanted to do fancy stuff with local systems then you could still build your own server the same way we have all built HTTP servers for decades and just expose the mcp.whatever.com subdomain for discovery. Then skip OIDC or allow ALL or something to simplify if you want.
Interesting read. Curious how the author feels re: the attack on airbases using shipping containers/drones that was so successful?
Seems to be a unique case that worked especially well for (higher end I'm sure) FPV drones. Getting artillery in on shipping containers would have a higher likelihood of detection. Similarly, the ability to 'guide' in the drones with munitions seemed to allow for greater flexibility during the attack and its effectiveness.
I imagine eventually these cheap FPV's will be augmented with low-cost GPU's allowing for running smallish models and self-guided autonomy. This would seem the next evolution where a commander deploys them in bulk and overwhelms the enemy in a way that can't be jammed like radio-communication. Similarly, horrifying when you consider their eventual use in terrorism scenarios...
That didn't use FPV drones, they're rather difficult to control at 6000km and they didn't have operators nearby.
Most likely it's the first major deployment of their semi autonomous drone tech, driven "declaratively". They've shown that stuff recently, they probably used it before showing it.
-- Design and develop machine learning models for various applications, including but not limited to transformers, hidden Markov models, GRUs, BERT, GPT, LSTMs, reinforcement learning models, and diffusion models.
-- Expertly preprocess and encode unstructured, time series, and structured data into suitable formats for diverse machine learning models.
-- Implement cutting-edge machine learning algorithms and frameworks to solve complex problems.
-- 4-5+ years of experience with strong academic background and excellent communication skills.
-- Full-time, on-site position with equity options and comprehensive health benefits.
Send resume and any relevant projects or links to frank (at) 8flow.com
So as someone who took many of those remedial math classes this article rubbed me a bit the wrong way. Not necessarily that any of his descriptions are wrong about the students or the type of things it covers. Instead there seemed to be this underlying theme in the article about it being a waste of time and of little value to society. I strongly feel any time us as humans sit in a classroom and try to better ourselves even when we ultimately fail, it benefits society overall.
Personally, I was in the basic math in High school, like long division/multiplication freshmen/sophomore year. When I first went to community college around the age of 16-19 I got farther, taking Algebra I and then II. However, once I reached Calculus I crashed and burned, although I did great in my CS and other science classes.
I eventually entered the work force (programming/tech) and over the next half dozen years tried and failed at least 3 times to restart at community college usually failing at Calculus I or college level english. Finally, at 30 it seemed to take, I eventually passed Calc I, Discrete, Linear, got my degree. A blend of community college and state school so I didn't break the bank.
I have friends from other sides of the world that have told me this would only really be possible in the US. In many places their is no equivalent of community/junior colleges or an attempt at adult remedial education. Instead you place in your teens, and if you score well enough you get to go to college. Otherwise, its trade school or similar and much more difficult to escape your socio-economic class. The author and others seem to be advocating for something similar here under the guise of it being unethical to waste resources or give hope to the dumb dumbs. I can't say I agree...
He's talking about taking students who can't do grade 7-9 math and charging them $200,000 and several years of their life for the promise of a career they almost never get. These students are getting screwed over hard. Most of them don't want to be taking the class.
I understand where you're coming from, the author definitely had a chip on his shoulder, but what he's describing is a situation where the students clearly do not want to be sitting in that classroom. They're being manipulated to believe they need to take out loans and go to college to succeed. That system just ends up wasting the time, money, and mental health of everyone involved.
-- Contribute to our AI-driven personalized automation platform.
-- Key skills: NodeJS, Python, Docker, Terraform, Serverless; GCP (Preferred); NoSQL/SQL databases.
-- 2-5+ years of experience in backend systems and cloud ecosystems.
-- Full-time, on-site position in Palo Alto.
Front-End Engineer, $120-140K/yr:
-- Proficiency in React, Typescript, Javascript, modern front-end development.
-- Experience delivering reactive UIs for web pages and Chrome Extensions (Manifest V3)
-- Skilled in transforming Figma mockups to live UIs.
-- Full-time, on-site position in Palo Alto.
Data Scientist/Machine Learning Engineer, $125-150K/yr:
-- Full-time contract role with potential to convert to employee.
-- Work on predictive models for user automation and workflows.
-- Experience with BigQuery, GCP ecosystem, and Vertex AI preferred.
-- Remote candidates considered, Palo Alto location preferred.
What we offer:
-- A dynamic environment for professional growth in a leading AI company
-- Competitive compensation + equity in a seed stage startup
-- An innovative and collaborative team culture
-- Medical, Dental, Vision
-- Comprehensive amenities including breakfast, lunch, dinner for on-site employees
Sorry no visa sponsorship at this time.
Interested in being a part of AI-driven innovation? Email your resume/LinkedIn and a brief introduction to frank at 8flow.com
So my $0.02 from being on both the hiring and trying to get hired side of the fence. Also I can't do a whiteboard interview to save my life, never could. I mostly think the process everyone is running makes a bit more sense for entry-level applicants. You are dealing with a candidate pool exactly like what you describe above. However, for anything above a junior dev it is horribly inefficient.
Instead I'm always surprised more places don't rely on references and prior experience. Yes people lie, but in my experience its relatively easy to tell the difference with a simple glance at their LinkedIn. If I look at a candidate that spent 2-3 yrs as a Software Engineer at some company that I'm relatively familiar with, and they seem to have 2nds and thirds to people I know in the industry, then pretty good chance they aren't lying. Same for people taking the time out to write recommendations for them. Even bigger signal if they want setup some calls with their prior co-workers who can vouch for them, to me that means they stand by their work and reputation.
I recently went through about seven rounds for a senior role. During that time I repeatedly offered to setup some time with my prior coworkers from those I directly managed, to peers, to those I reported to (executive team). My thinking being that they could hear from the horses mouth how I lead a team, my work ethic, etc. They did not take me up on the offer, which to me was crazy. Yes, I could be running some machiavellian scam with 2-3 people who also made a fake LinkedIn, I also could have put up fake articles in PR Newswire announcing my last position, could have spoofed all those blogs I co-authored from the company I worked at 2 jobs ago, etc. But really, wouldn't it be a better signal for a candidate to offer and have all these things?
Instead you see an industry that puts someone with ~10 yrs experience through a whiteboard interview. It makes no sense.
I think this is more in line with how hiring works for senior jobs. I've been part of 50+ hiring decisions and do it like this.
Social stuff is absolutely the first thing I'm checking. It's a quick test to spot total lies, e.g. you claim you worked somewhere but are connected with zero people on any social network, or claim attendance at a school with zero connection to anyone. If nothing else, it helps to build rapport for the interview.
I would 100%, absolutely trust a personal recommendation over a dumb whiteboard interview.
> Social stuff is absolutely the first thing I'm checking. It's a quick test to spot total lies, e.g. you claim you worked somewhere but are connected with zero people on any social network, or claim attendance at a school with zero connection to anyone. If nothing else, it helps to build rapport for the interview.
There exist quite a lot of people who have no account on any social network for privacy reasons.
> If I look at a candidate that spent 2-3 yrs as a Software Engineer at some company that I'm relatively familiar with, and they seem to have 2nds and thirds to people I know in the industry, then pretty good chance they aren't lying.
The problem is: there exist a lot of companies - you can only be familiar with a very small fraction of them. Also, for many big companies, the differences between departments or groups can be a lot larger than between companies of similar size and sector.
Honestly, I feel like this is another bi-product of 0% interest rates for so long. Bankers have essentially stopped caring at all about deposit amounts. Many banks still offer 0.x% interest even with most T-bills paying 5%+. It is an odd position from something that should be easy business, receive billions in deposits, invest in 6 month T-bills, carve off 0.5-1% of the profit amount for yourself, profit?
I'm convinced this is less about them actually losing money and more about them having sour grapes about not making 100x or something. Like the credit card is losing money when it is immensely popular and charges like 18% interest, HOW?
> Bigger banks are simply betting that people will not go through the trouble of moving their money.
This applies even at smaller banks that don't make savings account their primary product.
Like, my credit union offers 0.1 - 0.9% savings accounts depending on your balance. With a $250,000 balance, you qualify for some Premier savings account with 4.8%.
I said screw it. I created an account somewhere else that gives over 5%. Yeah, it means my money might take a couple days for me to get if it I absolutely need it, but I can't imagine a scenario where that would be a problem that my credit card can't take care of.
If you know SIEM, threat intel, logs, detection modeling/strategy, purple teaming, and adversary emulation — and you've been wanting to explore how LLMs/agents can accelerate detection engineering workflows — we'd love to talk.
The role A lot of it is what we call AI babysitting: guiding 1,000 kindergarteners (LLMs) as they try to become senior detection engineers. Sometimes they're brilliant, sometimes chaotic — your job is to shape that into something production-ready.
You'll be part detection engineer, part researcher, part builder, with a big say in shaping the product and how AI gets applied to real-world problems.
Details Full-time or part-time Remote-friendly (SF-based, local a plus but not required) Early team, high ownership, fast iteration
Email Resume: john at synthropic.com