AI Engineer: Agentic Systems & War Simulation

The goal is simple. Help India win the next war.

AlphaGo discovered moves in Go that no human had played in thousands of years of the game’s history. Moves that looked wrong to experts but proved devastatingly effective.

We’re building the equivalent for air warfare.

A system where AI agents play through ten thousand conflict scenarios overnight, discover strategies that human planners would never consider, and stress-test every plan against an adversary AI that actively tries to break it. India enters any conflict having already explored more of the strategic space than any human planning staff ever could. The war is fought in simulation a thousand times before it is fought once in reality, and the side that has done that preparation wins.

This is not a simulation with better graphics. It is a tool that out-thinks the adversary.

What You’d Build

You own the AI agent architecture. Someone else builds the simulation engine. You build the intelligence on top.

• Agents that reason about complex multi-day, multi-front operations and explore the strategy space at a scale no human planning staff can.
• Adversarial agents that find the weaknesses in any plan before the enemy does
• A learning loop where agents get meaningfully smarter over thousands of self-play iterations.
• A reasoning layer that can explain to a military commander why a strategy works and where it breaks, with specific evidence

The hard problems you’d spend your time on:

1. How do you represent the state of a complex, partially observable conflict so an agent can reason about it effectively?
2. How do you combine long-horizon strategic planning with tactical adaptation?
3. How do you build a Red agent that is genuinely dangerous rather than a scripted opponent?
4. How do you make agent reasoning auditable and trustworthy enough that a commander acts on it?

Why This Is Hard

The action space is enormous. The environment is partially observable and stochastic. The adversary adapts. Planning horizons span days, not moves. A single decision (when to commit reserves, which front to prioritize) cascades in ways that only become visible much later.

Standard approaches break in ways that are obvious once you understand the problem, but non-obvious if you’re pattern-matching from existing work. No standard playbook exists for this. You’d be designing the architecture from first principles.

Who This Is For

We don’t care about years of experience or seniority. We care about how you think.

The bar is: can you design and build an agentic architecture that outcompetes systems built by teams ten times your size at companies with a hundred times your budget? If yes, the rest doesn’t matter.

What we’re looking for:

• Exceptional first-principles thinking. You derive architectures from the problem, not from what you’ve seen others do.
• Deep independence. You take an ambiguous, open-ended problem and come back with a working system. No hand-holding, no waiting for specs.
• Technical brilliance in at least one of: reinforcement learning, LLM agent architectures, multi-agent systems, game AI. You don’t need to be broad. You need to be devastatingly good at the thing you’re good at.
• No defence background required. No specific tool or framework required.

"If the idea of building an AlphaGo for warfare doesn’t make you want to drop everything, this isn’t for you. If it does, we’d like to talk."