Product Engineer (Maritime Intelligence)

Intelligent Systems Group, Kawa Space 

Bangalore/New Delhi, India

The problem

In June 2026, a vessel in the Baltic dragged its anchor along the seafloor for roughly three hundred kilometres and cut two telecommunications cables. Investigators traced the furrow afterwards. Nobody watched it happen.

The satellites saw it. The SAR returns existed. What did not exist was anything that could hold it together into a picture coherent enough for a watch officer to look at, at three in the morning, and say: that ship is doing something that does not make sense, here are the four explanations, and here is the cheapest thing we could look at to tell them apart.

That screen does not exist anywhere in the world. You would build it.

The role: 

You are the second product engineer at O19E, and you own the entire surface between our world model and the human being who has to decide something.

The hard part is not the map. Anyone can draw a map. The hard part is designing for doubt — putting competing hypotheses, their evidence, their contradictions, the confidence attached to each, and the cost of resolving them in front of an analyst who has about three seconds. Almost every intelligence interface ever shipped projects false confidence, because confidence is easier to draw than uncertainty. Ours is not allowed to. That constraint is the job.

The second half of the job is the loop. When an analyst confirms, corrects or dismisses one of our assessments, that judgement has to be captured as a structured, governed label — not a free-text comment that dies in a database. Those labels are the only asset in this company that cannot be bought. The quality of the interface you build determines the quality of the data that trains everything else. Product is not downstream of the model here. It is upstream.

In 12 months you will have built:

• The Vessel Case File — timeline, identity-confidence history, network exposure, ranked hypotheses with evidence for and against, and the recommended next collection — in the hands of two or three design partners who use it on real cases.

• Vessel 360 — the full dossier for any hull, assembled from sensors, registries and behaviour, with every claim traceable to its source and timestamp.

• The adjudication loop — the capture surface that turns analyst judgement into training data, plus the instrumentation that tells us honestly whether case one hundred was faster and better-founded than case one.

The decision only you can make: how much uncertainty to show. Too little and we launder model confidence into false certainty, which violates the one doctrine we will not bend. Too much and no one can act, which makes us useless. Nobody else at the company can find that line. You will find it by sitting with analysts and watching them fail to use your first three attempts.

What you are walking into

We are early, and the honest version matters more than the flattering one.

We have a reference architecture we believe in, a thesis stress-tested against the primary literature, and a specific first wedge: dark-vessel and identity intelligence in the Indian Ocean corridor. Over the next six months we are running user interviews with defence teams of 4 countries, coast guard and sanctions analysts, standing up the data foundation, and hand-building twenty-five to fifty adjudicated case studies that become the seed corpus for everything after.

What we do not have: a signed anchor customer, labelled data at scale, or proof that our identity graph survives a determined adversary. You would be joining to help remove those three uncertainties, not after someone else did.

You will pick the stack. You will ship daily. You will talk to users directly, without a product manager in between, because at this stage there isn't one and shouldn't be.

You are also walking into this: https://www.kawaspace.com/philosophy
 

Who you will work with

You will work most closely with two people.

The maritime domain lead has stood watch or run a global intelligence cell, knows which alerts real analysts have learned to ignore, and will be adjudicating the first fifty cases by hand. This person is your fastest feedback loop and your reality check. When your elegant abstraction quietly stops matching how the work is actually done, they are the one who will tell you, and they will be right.

The graph and identity engineer owns the hardest problem in the domain — a temporal identity graph that holds competing hypotheses, tolerates identifier reuse, and can always explain why it ranked one linkage above another. Your interface is where their probabilistic output either becomes legible or becomes noise. You two will argue often. That is the point.

The third person you’ll spend a bit of time with is me, the founder. 

What we actually require

There are a few hard requirements.

• You have shipped a product end to end, to real users, and iterated on it after it hurt. If you are a fresher with Claude code slop please do not apply (please do not DM.)  

• You are comfortable working where the domain is genuinely unfamiliar and the right answer is unknown to everyone in the room, including us.

• You can work with sovereign and government customers where the role requires it. 

• You don’t post slop on Linkedin, or write crap like “humbled/excited to announce.” 

• Culture fit. I know what I'm looking for. 

Everything below is curiosity, not a filter. Please apply if some of it is true and none of it is:

• Geospatial, map or time-series interfaces at scale

• Real-time or streaming data surfaces

• Any exposure to defence, intelligence, maritime, aviation or emergency response

• Design instinct strong enough that you have shipped without a designer, or fought a designer and won on evidence

• Experience being the first product hire somewhere, and knowing what that actually costs

• A gruelling degree in core sciences. (Data Science is not Science. ‘Social sciences’ is not sciences) 

We are not filtering on years, framework familiarity, or a defence background. The best person for this role may well come from a field that has nothing to do with the ocean.

The offer

• 24-36L per annum, depending on what you shipped. 

• A nice small boat for you to ride, when you are in Mumbai. 

Beyond that, three things this role gives you that most cannot:

Access. You will sit with global naval intelligence analysts, coast guard watch officers and sanctions investigators, and watch them work. Very few product engineers ever get to see the actual moment of decision they are designing for.

Ownership with teeth. You own the product surface and the adjudication loop, which means you own the quality of the data that trains every model we build. That is not a scoped mandate. It is a structural one.

Work you can show. We intend to publish an open benchmark for our product, including the tests we fail. Your evaluation work on interface calibration — whether people's trust in our output actually tracks its accuracy — is publishable, and we want it published.

Alright, 

Tell us about something you shipped that was honest about its own limitations, and what that cost you. 

Apply w. a letter (pick a theme/challenge from this JD and tell me how’d solve it). Apply below.