Starboard Maritime Intelligence | Agentic AI Engineer | Wellington, New Zealand | Full-time | On-Site
Starboard builds mission-critical intelligence platforms for the defence, national security, and maritime enforcement sectors. Our software fuses complex datasets—including satellite imagery (SAR/EO), vessel tracking (AIS), and proprietary intelligence to help our customers identify and respond to illicit and threatening activities at sea. We work on challenges like sanctions evasion, illegal fishing, and maritime security operations.
We are seeking our first Agentic AI Engineer to pioneer a new capability within the company. This is a foundational role with significant autonomy, responsible for designing and building an "AI Agent Workforce" from the ground up. Your objective will be to leverage agentic AI to enhance operational effectiveness and accelerate the intelligence cycle across our business—from engineering to analysis. This is a hands-on-keyboard role focused on deploying robust, auditable, and secure AI systems, not just theoretical research.
Example responsibilities:
- Develop agents to automate the initial analysis of raw data streams, generating draft intelligence summaries for our analyst teams.
- Build agents to assist with static code analysis, security vulnerability scanning, or triaging failures in our secure deployment pipelines.
- Create agents to perform real-time horizon scanning, monitoring multiple data feeds to flag and contextualise key maritime events.
- Establish the company-wide standards, tooling, and security frameworks for the development and deployment of all AI agents.
We're looking for someone with:
- Demonstrable, hands-on experience building functional AI agent systems. This can be from professional, academic, or personal projects.
- Proficiency in Python.
- A deep-seated curiosity for emerging AI technologies and a systematic approach to experimentation.
- An understanding of how to apply technology to solve concrete business and operational problems, with a strong sense of ownership.
IMPORTANT - SECURITY REQUIREMENTS: Due to the nature of our work with government defence and intelligence partners, this role has strict security clearance requirements. To be eligible, you MUST have been a citizen of the UK, USA, Canada, New Zealand, or Australia for at least 10 years. We regret that we cannot make exceptions to this.
If you are interested in applying your AI skills to critical national security and intelligence challenges, we would love to hear from you.
For those interested in a high-level walk through of the Anglo-Saxon period, I highly recommend “The Anglo-Saxons: A History of the Beginnings of England” by Marc Morris. It covers the period from a number of different perspectives, including a clerical/monastic one that is easy to underestimate and ignore with a modern secular bias.
There is also a very listen-able audiobook version available on Audible.
It’s used by Mistral, AWS, Cloudflare, and countless others.
vLLM, HF TGI, Rayserve, etc are certainly viable but Triton has many truly unique and very powerful features (not to mention performance).
100k DAU doesn’t mean much, you’d need to get a better understanding of the application, input tokens, generated output tokens, request rates, peaks, etc not to mention required time to first token, tokens per second, etc.
Anyway, the point is Triton is just about the only thing out there for use in this general range and up.
You can Google all of them + nvidia triton, but here you go...
Mistral[0]:
"Acknowledgement
We are grateful to NVIDIA for supporting us in integrating TensorRT-LLM and Triton and working alongside us to make a sparse mixture of experts compatible with TRT-LLM."
Cloudflare[1]:
"It will also feature NVIDIA’s full stack inference software —including NVIDIA TensorRT-LLM and NVIDIA Triton Inference server — to further accelerate performance of AI applications, including large language models."
Amazon[2]:
"Amazon uses the Text-To-Text Transfer Transformer (T5) natural language processing (NLP) model for spelling correction. To accelerate text correction, they leverage NVIDIA AI inference software, including NVIDIA Triton™ Inference Server, and NVIDIA® TensorRT™, an SDK for high performance deep learning inference."
There are many, many more results for AWS (internally and for customers) with plenty of "case studies", and "customer success stories", etc describing deployments. You can also find large enterprises like Siemens, etc using Triton internally and embedded/deployed within products. Triton also runs on the embedded Jetson series of hardware and there are all kinds of large entities doing edge/hybrid inference with this approach.
You can also add at least Phind, Perplexity, and Databricks to the list. These are just the public ones, look at a high scale production deployment of ML/AI in any use case and there is a very good chance there's Triton in there.
I encourage you to do your own research because the advantages/differences are too many to list. Triton can do everything you listed and often better (especially quantization) but off the top of my head:
- Support for the kserve API for model management. Triton can load/reload/unload models dynamically while running, including model versioning and config params to allow clients to specify model version, require specification of version, or default to latest, etc.
- Built in integration and support for S3 and other object stores for model management that in conjunction with the kserve API means you can hit the Triton API and just tell it to grab model X version Y and it will be running in seconds. Think of what this means when you have thousands of Triton instances throughout core, edge, K8s, etc, etc... Like Cloudflare.
- Multiple backend support with support for literally any model: TF, Torch, ONNX, etc with dynamic runtime compilation for TensorRT (with caching and int8 calibration if you want it), OpenVINO, etc acceleration. You can run any LLM (or multiple), Whisper, Stable Diffusion, sentence embeddings, image classification, and literally any model on the same instance (or whatever) because at the fundamental level Triton was designed for multiple backends, multiple models, and multiple versions. It operates on a in/out concept with tensors or arbitrary data. Which can be combined with the Python and model ensemble support to do anything...
- Python backend. Triton can do pre/post-processing in the framework for things like tokenizers and decoders. With ensemble you can arbitrary chain together inputs/outputs from any number of models/encoders/decoders/custom pre-processing/post-processing/etc. You can also, of course, build your own backends to do anything you need to do that can't be done with included backends or when performance is critical.
- Extremely fine grained control for dispatching, memory management, scheduling, etc. For example the dynamic batcher can be configured with all kinds of latency guarantees (configured in nanoseconds) to balance request latency vs optimal max batch size while taking into account node GPU+CPU availability across any number of GPUs and/or CPU threads on a per-model basis. It also supports loading of arbitrary models to CPU, which can come in handy for acceleration of models that can run well on unused CPU resources - things like image classification/object detection. With ONNX and OpenVINO it's surprisingly useful. This can be configured on a per model basis, with a variety of scheduling/thread/etc options.
- OpenTelemetry (not that special) and Prometheus metrics. Prometheus will drill down to an absurd level of detail with not only request details but also the hardware itself (including temperature, power, etc).
- Support for Model Navigator[3] and Performance Analyzer[4]. These tools are on a completely different level... They will take any arbitrary model, export it to a package, and allow you to define any number of arbitrary metrics to target a runtime format and model configuration so you can do things like:
- p95 of time to first token: X
- While achieving X RPS
- While keeping power utilization below X
They will take the exported model and dynamically deploy the package to a triton instance running on your actual inference serving hardware, then generate requests to meet your SLAs to come up with the optimal model configuration. You even get exported metrics and pretty reports for every configuration used/attempted. You can take the same exported package, change the SLA params, and it will automatically re-generate the configuration for you.
- Performance on a completely different level. TensorRT-LLM especially is extremely new and very early but already at high scale you can start to see > 10k RPS on a single node.
- gRPC support. Especially when using pre/post processing, ensemble, etc you can configure clients programmatically to use the individual models or the ensemble chain (as one example). This opens up a very wide range of powerful architecture options that simply aren't available anywhere else. gRPC could probably be thought of as AsyncLLMEngine on steroids, it can abstract actual input/output or expose raw in/out so models, tokenizers, decoders, clients, etc can send/receive raw data/numpy/tensors.
- DALI support[5]. Combined with everything above, you can add DALI in the processing chain to do things like take input image/audio/etc, copy to GPU once, GPU accelerate scaling/conversion/resampling/whatever, pipe through whatever you want (all on GPU), and get output back to the network with a single CPU copy for in/out.
vLLM and HF TGI are very cool and I use them in certain cases. The fact you can give them a HF model and they just fire up with a single command and offer good performance is very impressive but there are an untold number of reasons these providers use Triton. It's in a class of its own.
That is simply not correct. With properly implemented E2EE, you can communicate confidently over a completely insecure channel. You could post the entire data stream publicly on the internet with no loss of privacy.
I would still assume that with apps like Snapchat, there are inherent problems on meta-level (from an E2EE view) because for the users, Snapchat is essentially a trusted third party, providing governance/management features. But it can well be that Snapchat does not provide E2EE; I cannot see any remarks about it on their webpage.
"WhatsApp _considers_ chats with businesses that use the WhatsApp Business app or manage and store customer messages themselves to be end-to-end encrypted."
And later, they write that in many cases, Meta can actually read the messages.
> but those operations over what usually appears at the denominator of the derivative purely for notation purpose has always looked to me as complete magical garbage.
You will be delighted to discover that they are in fact not magical or garbage.
> What object is "dx" ? is it a number ? a limit ? is it zero ? can i divide another number by it ?
dx is a differential one-form. You can think of it as a generalisation of a gradient, if you like. These are very important in Differential Geometry.
You can use differential forms to do all sorts of things, but one example you may be familiar with is to compute area or volume forms over arbitrary manifolds. It gets a bit hard to define things on HN without TeX support, but using differential one-forms and the related exterior derivative, you can define a generalised Stokes' theorem that works for any smooth, oriented manifold.
I used this in my PhD, and implemented it directly in a numerical method, so this has very practical engineering uses also.
"Elementary Differential Geometry" by Barrett O'Neill is a pretty beginner-friendly introduction to some of these topics if you're interested, though there are many other good texts also.
>dx is a differential one-form. You can think of it as a generalisation of a gradient, if you like. These are very important in Differential Geometry.
This really doesn't help beginners. At all.
There are formal contexts where we can reinterpret division by zero and have it make sense. Should I start telling students that division by zero is allowed? Should I start teaching intro calculus students that 1+2+3+...=-1/12?
For teaching purposes you are definitely allowed to lie, as long as that lie can be resolved eventually (not necessarily in this semester ;-). That's how we have been generally taught about integer divisions and negative square roots. But behind the scene, the `dx` notation can be fully generalized and made rigorous with differential forms, or that was what I have been told.
This is definitely not an apples to apples comparison. Integer division is something everybody is expected to learn. Also we don't teach imaginary numbers to middle schoolers as soon as they learn about square roots.
To some extent we have to speak to our audience. I consider that part of effective communication. I don't think "assume the person you're speaking to is/will be a mathematician" is an effective way to interact.
I meant that, yes you are right. You are not expected to teach differential forms to non-math students at all because it's not effective. The existence of differential forms only means that it can be eventually made rigorous if you push hard.
dx can be a differential form, but in (elementary) calculus books it's exposited this way: Suppose you have a function y = f(x) taking real numbers to real numbers, so x is an independent variable, y is the dependent variable. You define an independent variable Δx and define dx = Δx. Also define dependent variables dy and Δy by
Δy = f(x + Δx) - f(x) and dy = f'(x) dx.
Then f'(x) = dy/dx. This may look like a stupid hack to make the last formula work, but actually it's a little more. If you use nonstandard analysis, you define the derivative of a function f from reals to reals by
f'(a) = st( (f(a + Δx) - f(a)) / Δx )
where st takes the standard part of a hyperreal number and Δx is a nonzero infinitesimal. This is like the usual limit definition, without limits. Then you can use the formulas above and "dy" and "dx" are numbers, albeit hyperreal numbers.
(The "dx as a differential form" vs. "dx as a number" is probably coming from the fact that the tangent space to the reals at a real number is isomorphic to the reals, so the dual space [where dx lives] is too.)
(Calculus via infinitesimals is pretty cool; a good resource for this is H. Jerome Keisler's "Elementary Calculus" and "Foundations of Infinitesimal Calculus", both available for free: https://people.math.wisc.edu/~hkeisler/)
I second the recommendation for Barrett O'Neill's book - I used it in my differential geometry class at MIT.
As the father of two young daughters, I adore Bluey. Quite apart from how good the episodes are for adults, I really enjoy the following aspects:
1. It portrays wholesome and positive family relationships, particularly between father and daughters. The girls are also good to one another, with occasional conflict kindly and realistically resolved.
2. It tells simple stories without pushing political messages. Unfortunately, children’s television at large has become another victim of the culture wars, with heavy-handed social messages infecting what should be simple cartoons. Bluey is something I can trust.
3. I don’t have to expose my kids to American accents. They’ll have more than enough of that from regular programming when they grow up.
> "This family has two mommies. They love each other so proudly and they all go marching in... the... big parade," sing the lyrics. Other terms featured include "trans," "non-binary" and "queer."
> “Love is love is love you see, and everyone should love proudly,”
Acknowledging LGBTQ people are capable of love is somehow an example of “culture war”. It’s incredible people can be so bothered by a simple, sweet message.
yes, acceptance of others is an overly political message to those of a certain political persuasion. can’t have kids running around thinking whatever they grow into might be ok if it doesn’t comport with mommy and daddy’s worldview.
Whatever kids grow up into is not okay with most parents. I’d imagine almost everyone would agree with that statement. There are probably traits you wouldn’t want your children to have. You probably wouldn’t be okay with them being gambling or drug addicts, racist, murderers, uneducated, etc.
If parents were truly okay with their children growing up into whatever they happen to grow into, there would be no reason for the parents to impart morals, life lessons, education, proper nutrition, etc., to children. The fact that, for the entire existence of humankind (so far as I can tell), we seem to have done those things, it seems to me that all parents strongly care about what their children grow up to be like.
Why does the idea of a parent wanting to impart their vision of how the world ought to be offend you? If the shoe were on the other foot, and you wanted your children to be raised a particular way, would you appreciate being ridiculed for that?
>You probably wouldn’t be okay with them being gambling or drug addicts, racist, murderers, uneducated, etc.
Some are objective. Not much interpretation to whaty a gambler or drug addict is. But the sad thing is that "racism" and especially "educated" are highly subjective.
>Why does the idea of a parent wanting to impart their vision of how the world ought to be offend you?
Depends on the vision. Obviously as a black man I'd have an issue with people who's "vision" is that black people are dangerous, dumb, and dirty. I imagine it's the exact same for a lesbian, or a trans person, or even a not rich person.
And of course not all disagreements are equal. I may not prefer a helicopter approach but that is ultimately a choice that does not impact me nor my family. The former, not so much. Your freedom ends where mine begins.
It goes both ways. Maybe you'll teach your kids to be open and tolerant to everything but they'll grow up, convert to Catholicism, and -- gasp -- vote conservative.
Yes, i fully recognize that at least one of my children is likely to enjoy baseball and vote for a Republican at least once in their life, despite my best efforts to show them the light. I will love them anyway.
well, everything is political, but i wasn’t making it so. i just really hate baseball and have never encouraged my kids to play the sport because of that.
Perhaps you mean 'vote reactionary', as the religious tend to do? Voting conservative means the Democratic party these days - general support of institutions, the United States's standing in the world, law and order, individual freedoms, fiscal responsibility. The contemporary Republican party has rejected these things, instead focusing on some imagined idyllic past and pushing for radical change towards it. The Democratic party certainly has its things it wants to change, but they're much more incremental and not the sweeping sea change of the current Republican agenda.
Bluey is targeted towards pre-school children. I'm not really bothered if my 4-year-old daughter sees stuff like this, it's just that she won't really understand things like "non-binary". It seems like a topic more appropriate for older children - maybe from age 8 or later?
At my kids’ childcare, by far the most popular educator is non-binary. They do a great job of listening to the kids, and talking to them without talking down to them. Come to think of it, the way they interact with the kids actually reminds me a fair bit of the parents in Bluey.
Both my kids (3 and 5 but now at school) have said stuff like “Today we did x with Greg. He’s not exactly a boy and not exactly a girl.” Then they get on with their day. To them, it’s just another person that’s a bit different to them.
I live in a pretty liberal European city, yet I haven't met any non-binary person. I know exactly one trans person, and only remotely (she lives in US). I guess to me, this topic seems "advanced", perhaps "irrelevant" in a way for such a small kid. There are many other things she needs to learn about, which she will commonly experience in the real world.
80s in California... there was a sizable but not exactly large group. No one cared. We all knew "old people" were supposed to care, but it was more likely just a thing everyone had to pretend to care about on TV. None of the old people I knew cared either.
Depends on the circle. Tech industry in California, know 2 NBs, and 2 trans people (and then 2 more I knew transitioned after I left work and contact with them). I guess that does fit all the liberal stereotypes that people like to throw at my State.
I'm in London, and I've met a few non-binary people (enbys), and had at least two trans people in my school (one student, one teaching assistant; this was in the mid-2000s).
> Some people believe that whether you're a woman or a man is a thought in your head, and they also believe that these thoughts mean you can be neither, which they call 'non-binary'. It's helpful to respect these people's beliefs and act as if they are true, because they can get very angry and vindictive if you don't agree with them.
And then we ask why some kids wind up entrenched away from economic opportunity…
Like, if your only coping mechanisms for beliefs you disagree with—particularly about someone else’s private affairs—rise out of fear of retribution, you shouldn’t be in a decision-making role of any kind. It’s somewhat sad to see that baked into a kid from the get go, but maybe they’ll get over it without winding up resentful for the handicap.
The point is it's not a private affair. Retribution from people who react harmfully when others do not share their beliefs is a real thing, ask anyone who was brought up in a strict religious environment who became a non-believer. Sometimes the easiest path is minimal appeasement to avoid conflict where you'll end up worse off.
Someone's sexuality or gender identity sure is. Given languages' pronouns evolve (e.g. a universal "you" in place of the informal "thou", or the aborted deprecation of "y'all"), that's not a reasonable hang-up.
> the easiest path is minimal appeasement to avoid conflict where you'll end up worse off
Sure, and instinctive conflict avoidance is a valid life strategy. It's just bad build for a decision maker. Someone conditioned in that behaviour is going into life with opportunities cordoned off.
Perhaps you should teach children that it's good to respect people's beliefs because being nice to people is good?
I think it's ironically tragic that in an attempt to get people to respect your beliefs more, you argue that the main reason to do so it out of fear of retribution.
The one thing that could be noted on this point of seeing a toddler-focussed TV show with an LGBTQ acceptance theme, is that there are a lot of other acceptance-worthy themes out there that never make it into toddlers' TV shows. I will not make a list because obviously that would be whataboutism, virtue signalling, I don't know.
That too - the pantomime dame character is basically just men ridiculing middle-age women, while sending a message to any children watching that it's acceptable to do so. It's as offensively sexist as the drag queens.
If I'd written a comment about the racism inherent in 'blackface' performance regarding its mockery of black people, would you claim I was redefining racism?
Yes, you did. Both blackface and drag performance have particular histories you're intentionally and erroneously conflating. The former is racist in its origin. The latter is not misogynistic in either its origin or contemporaneous performance.
Drag is men dressing up as caricatures of women. They wear costumes intended to represent women and mimic female bodies, adopt a 'woman' persona under a feminised and often heavily sexualised name, and act out every demeaning, offensive stereotype of women for laughs, often leaning heavily on mocking women's bodies and the physical experiences exclusive to women: pregnancy, birth, breastfeeding, menstruation, and sometimes even abortion. Then they take off those costumes and get to go about their lives as men when they aren't doing this, without having to live under those same stereotypes they helped perpetuate for fun or money. Meanwhile, women are expected to laugh, clap along and celebrate this insult. This mockery of women isn't exactly subtle.
So, please explain the reasoning behind your belief that drag is not misogynistic.
I guess you think that bodybuilders, models, and actors (film and especially Theatre) are all misandric/misogynistic as well? They fit all the items above as well.
I guess I see why the comment upstream was flagged.
Please explain your logic more clearly, I don't see the connection you're making between the occupations you mentioned and men dressing up to make an offensive mockery of women in the way they do for drag.
>They wear costumes intended to represent [Person]
>adopt a persona under Fake name
>act out every demeaning, offensive stereotype for s/laughs/entertainment, boften leaning heavily on the physical experiences.
>Then they take off those costumes and get to go about their lives, without having to live under those same stereotypes they helped perpetuate for fun or money
How is RuPaul doing anything different that America's next top model, Hollywood, or any other competition based on looks doing? It's just personal interpretations if you view it as empowering, demeaning, or even bigoted. They all get the same accusations levied at them after all.
How is this different to when Mr. Rogers shared a wading pool with his Black mailman? Or Captain Kirk kissed Uhura? Both evoked the same conservative outrage we hear about today over things like this. Introducing and celebrating people different from ourselves is how we become more comfortable with them. It eliminates the other-ness of them. As these kids grow up they're going to become aware of a vocal minority of adults who are trying to convince people to marginalize these minority groups and in some cases even call for their extermination. Yes it's a culture war, but it's not being waged by LGBT people, it's being waged against them.
To be fair, there are subtler ways of communicating acceptance. The Bluey way would seem to be having a couple at the playground that the kids maybe ask about. Or maybe they don’t! They just coëxist peacefully because it’s no more remarkable than them having a different skin tone. Different framing from a drag show, which I can see someone preferring without being homophobic.
That's because its never heavy handed.. How about the episode with grandpa struggling with the new app centric lifestyle: Phones "Are you sure we're playing this right?". Subtly provocative and brilliantly done.
> 3. I don’t have to expose my kids to American accents. They’ll have more than enough of that from regular programming when they grow up.
What’s the concern here? I understand American media dominates everything, so I’m guessing you don’t want your kids to pick up small bits of American accent when they’re still developing?
I'm not the OP, but as a child growing up in an anglophone country but with a 'low-prestige' accent (think Indian English or Singaporean English), it was ingrained to me pretty early on that the only 'proper' ways to speak were RP or General American, and one of the reasons was because all the kid shows I watched had characters who talked almost exclusively in those accents. 'Exotic' accents were used mostly for comedic value.
Having a greater variety of shows that model the diversity of the English language would probably have gone some way to dispelling that damaging self-belief? I really love the existence of shows like Derry Girls, because they make me less self-conscious of my 'exotic' accent.
That's a really interesting perspective that I hadn't considered, thank you for answering! I have what anyone would call a General American accent, so your experience isn't one I'd ever been personally exposed to growing up. I remember a lot of older sitcoms and cartoons using "exotic" accents for comedic value though, like you said, and it's obvious that can be damaging to anyone's self-esteem.
>2. It tells simple stories without pushing political messages.
I did wonder if there'd be any outcry over the DIY episode which is about evolution. Certain cultures seem to have serious issues with that. It did also hint at afterlife, however.
I am entirely on Meta's side here. They have done a huge amount for open source ML, and with the llama2 license they are enabling millions of small-to-very-large companies with incredibly valuable technology, while asking the hyperscale cloud providers to pay. This is more than reasonable, in my view.
I would like to see more open source technologies taking this kind of licensing view - free for almost everyone, except the frankly parasitic hyperscalers who profit from others' largesse.
Starboard builds mission-critical intelligence platforms for the defence, national security, and maritime enforcement sectors. Our software fuses complex datasets—including satellite imagery (SAR/EO), vessel tracking (AIS), and proprietary intelligence to help our customers identify and respond to illicit and threatening activities at sea. We work on challenges like sanctions evasion, illegal fishing, and maritime security operations.
We are seeking our first Agentic AI Engineer to pioneer a new capability within the company. This is a foundational role with significant autonomy, responsible for designing and building an "AI Agent Workforce" from the ground up. Your objective will be to leverage agentic AI to enhance operational effectiveness and accelerate the intelligence cycle across our business—from engineering to analysis. This is a hands-on-keyboard role focused on deploying robust, auditable, and secure AI systems, not just theoretical research.
Example responsibilities:
- Develop agents to automate the initial analysis of raw data streams, generating draft intelligence summaries for our analyst teams.
- Build agents to assist with static code analysis, security vulnerability scanning, or triaging failures in our secure deployment pipelines.
- Create agents to perform real-time horizon scanning, monitoring multiple data feeds to flag and contextualise key maritime events.
- Establish the company-wide standards, tooling, and security frameworks for the development and deployment of all AI agents.
We're looking for someone with:
- Demonstrable, hands-on experience building functional AI agent systems. This can be from professional, academic, or personal projects.
- Proficiency in Python.
- A deep-seated curiosity for emerging AI technologies and a systematic approach to experimentation.
- An understanding of how to apply technology to solve concrete business and operational problems, with a strong sense of ownership.
IMPORTANT - SECURITY REQUIREMENTS: Due to the nature of our work with government defence and intelligence partners, this role has strict security clearance requirements. To be eligible, you MUST have been a citizen of the UK, USA, Canada, New Zealand, or Australia for at least 10 years. We regret that we cannot make exceptions to this.
If you are interested in applying your AI skills to critical national security and intelligence challenges, we would love to hear from you.
Apply here: https://www.seek.co.nz/job/85965176