Website:
smartsystemofsystems.com
Job details:
Build the spine of the decision graph — the distributed system that ingests heterogeneous sensor streams under contested bandwidth and composes them into one coherent picture.
The mission
Our decision graph is the connective tissue of the company. It receives multi-modal sensor streams from radar, EO/IR, SIGINT, and acoustic platforms — composes them into a single decision-grade picture — and continues to do so when the upstream link drops, when nodes go dark, when timing is adversarial.
You will own its architecture from first principles.
This is a role for someone who has been in the room for hard distributed-systems decisions before, and who is ready to make them again — this time with the consequences of getting it wrong measured in lives, not lost analytics events.
A PROBLEM YOU'D JOIN US IN SOLVING
"Radar at 50 Hz, EO/IR at 25 fps, SIGINT pulses arriving asynchronously — fused on a 4 GB Jetson, with the upstream link dropping for 90 seconds. Operators downstream still need a decision-grade picture every second. How do you architect the buffering, the fusion, and the graceful downgrade?"
You'll own 5 SURFACES
- Core fusion pipeline architecture — protocol design, state replication, consistency model.
- Edge ↔ cloud handoff across degraded and intermittent links.
- Observability and replay tooling — production incidents must be reconstructable.
- Performance budget across all sensor ingestion paths.
- Mentoring the next two engineers on this team.
You bring 7+ YEARS
- Production distributed systems in Rust, Go, or modern C++.
- Deep familiarity with consensus, eventual consistency, CRDTs, and their trade-offs.
- Comfort with low-level networking — UDP, multicast, custom protocols on lossy links.
- A bias for systems that fail gracefully — degraded beats silent.
- Habit of writing technical specs, not just code.
Bonus signals OPTIONAL
- Background in defence, aerospace, robotics, or autonomous vehicles.
- Familiarity with tactical radio links (Link 16, MIDS) or analogous lossy military networks.
- Open-source contributions to ZeroMQ, NATS, or a real-time framework.
Take multi-modal model architecture from research into production — small, fast, contested-edge models that coordinate with each other and degrade gracefully under adversarial conditions.
The mission
Our autonomy stack lives on platforms that have 4 GB of RAM, 25 W of thermal headroom, and a connection that drops every few minutes.
You will design models that ship in that environment — not papers in a notebook, models in the field.
You'll make architectural decisions about on-platform inference, multi-agent coordination, and adversarial robustness. You will care about the gap between IID benchmark performance and what happens when sensors get spoofed, jammed, or simply broken.
A PROBLEM YOU'D JOIN US IN SOLVING
"Twelve heterogeneous platforms — four fixed-wing, six quadcopter, two ground — each carrying its own sensor mix. Maintain coverage of a 50 km² area while one platform is lost every three minutes. No GPS. Limited inter-platform comms. How do you train it, and how do you know it works?"
You'll own 5 SURFACES
- Model architecture for on-platform inference under tight memory and power budgets.
- Multi-agent coordination — RL, learned policies, or hybrid classical/ML approaches.
- Adversarial robustness — deception, jamming, sensor spoofing.
- The training-to-deployment pipeline for sovereign environments — no cloud GPUs at customers.
- Field-relevant evaluation harness — benchmarks that predict real performance.
You bring 5+ YEARS
- Shipped ML in production — not just papers, deployed.
- Deep PyTorch + ONNX/TensorRT or equivalent edge runtime experience.
- Strong intuition for the gap between benchmark and field performance.
- Familiarity with RL, multi-agent learning, or operations-research optimisation.
- Willingness to write CUDA/Triton when it matters.
Bonus signals OPTIONAL
- Prior perception work for autonomous vehicles, robotics, or aerospace.
- Experience designing red-team evaluations.
- NeurIPS / ICLR / RSS publications, or shipped to a defence customer.
Make the system actually run. On real hardware, in the field, in vehicles and on platforms designed two decades ago. Own the edge runtime — kernel, drivers, memory, thermals, and the discipline of compute under budget.
The mission
Software people love to talk about edge. You will deliver edge — bring up new compute targets, debug timing on a CAN bus that hasn't been touched since 2008, hold a soldering iron, and ship a runtime that survives 55 °C in the back of a vehicle.
This is a role for someone who has scars — who has fixed silent failures at 03:00, who has read errata documents end-to-end, who knows that "it works on my desk" is a beginning, not an end.
A PROBLEM YOU'D JOIN US IN SOLVING
"A platform integrator hands you a 10-year-old vehicle bus with proprietary timing, and asks you to land a 2 GB model plus our fusion runtime on a Jetson Orin NX with a 25 W thermal budget — sustained at 55 °C ambient. You have eight weeks. What's the plan?"
You'll own 5 SURFACES
- The edge runtime hosting our distributed graph and inference workloads.
- Bring-up for new compute targets — Jetson, Snapdragon, AMD embedded, custom SoMs.
- Power, thermal, and ruggedisation work alongside platform OEMs.
- Low-level I/O — sensor SDKs, GPIO, serial buses, time sync (PTP).
- The "it doesn't work in the truck" debugging culture.
You bring 7+ YEARS
- Embedded Linux or RTOS (Zephyr, FreeRTOS) in production.
- Strong C / C++ / Rust on bare metal, with healthy fear of UB.
- Yocto / Buildroot / OE-core experience.
- Deep familiarity with at least two SoC families and their quirks.
- Field debugging instincts — JTAG, traces, oscilloscope when needed.
Bonus signals OPTIONAL
- Rugged or defence platforms (MIL-STD-810, DO-178).
- Linux kernel or major BSP contributions.
- Hardening embedded systems against tampering and side-channels.
Build the interface operators actually use — under stress, in degraded conditions, with consequences that matter. Operators receive options, not dashboards. You will own how those options look, explain themselves, and keep a human as the decision authority.
The mission
An operator at 02:00, in a darkened room, with three sensor feeds disagreeing and 1.2 seconds to commit — the screen they look at is your work.
You will design and build it: the typography, the information architecture, the motion, the moment a recommendation surfaces with its provenance and confidence intact.
This role is for a designer who codes — or an engineer with real design taste. You will spend time with operators in the field. You will own the design system, the front-end stack, and the discipline of "informationally dense and still calm."
A PROBLEM YOU'D JOIN US IN SOLVING
"Three sensor streams disagree. The system has 1.2 seconds to surface options to an operator who must commit. What does that screen look like, what does the operator see first, and how do they understand which option to pick — without becoming a passive forwarder of machine verdicts?"
You'll own 5 SURFACES
- The operator surface across web, native, and embedded operator stations.
- Information architecture for "options with provenance" — confidence and counterfactual on every recommendation.
- Real-time decision-graph visualisation — legible at 02:00 in a darkened room.
- The shared design system across all customer surfaces.
- Direct field work with operators — they are your second supervisors.
You bring 5+ YEARS
- A designer who codes, or an engineer with a real design eye.
- Production TypeScript + React + Canvas/WebGL or equivalent.
- Strong taste — you can tell when something is dense and still calm.
- Experience designing for high-consequence environments — medical, aviation, finance ops, defence.
- A portfolio of shipped work, not just shipped features.
Bonus signals OPTIONAL
- Monitoring dashboards, control rooms, or operator tooling.
- Map-heavy UIs — Mapbox, Cesium, Deck.gl.
- Comfort with Figma + handoff that doesn't lose nuance.
Be the bridge between engineering and Indian defence customers. Translate operational doctrine into product requirements; translate engineering reality into honest customer conversations. Help us win programmes that matter — and decline programmes that don't.
The mission
Defence procurement rewards the company that
understands the customer's problem better than the customer can express it. You will be the person who does that understanding — sitting in MoD reviews, walking platforms with primes, returning to engineering with translated, prioritised, honest requirements.
This is a senior role for someone who has been in or alongside Indian defence for a decade or more — someone who has seen the integrator landscape from the inside, who has walked away from a deal that wasn't right, and who can do it again.
A PROBLEM YOU'D JOIN US IN SOLVING
"A flagship customer wants a feature that takes six months and serves precisely one programme. The roadmap already has a competing item that serves three programmes and aligns with our core architecture. The customer is the second-largest in the company. Walk me through the conversation you'd have — internally and externally."
You'll own 5 SURFACES
- Full lifecycle of an Indian defence engagement — RFI, RFP, evaluation, deployment, sustainment.
- Customer trust as the default state — managed by the calendar, not by spin.
- The translation layer between operator language and engineering language.
- Joint working sessions with platform integrators and primes.
- The honest no — to engineering when something isn't working, to customers when something isn't ready.
You bring 10+ YEARS
- Indian defence, aerospace, or tier-1 system integrator background.
- Deep familiarity with MoD procurement, DPSU dynamics, and the integrator landscape.
- Operator empathy — preferably from time in uniform or alongside it.
- Comfort with engineers who push back, and the discipline to push back yourself.
- A strong moral compass — you've walked away from deals before.
Bonus signals OPTIONAL
- Service or veteran background — Army, Navy, Air Force, Coast Guard.
- Prior work at HAL, BEL, BDL, DRDO, or a private prime.
- Defence startup experience — not just defence prime experience.
Click on Apply to know more.
