About the Role

We're building a next-generation Multi-INT (Multiple Intelligence) Search & Reasoning Platform for the defense and intelligence sector. This is frontier-level AI engineering: designing reasoning architectures, building advanced retrieval systems, and orchestrating multi-model pipelines over millions of unstructured intelligence documents.

We're looking for an exceptional Applied AI Engineer to design and build the core AI system: the reasoning engine, retrieval pipeline, and multi-agent orchestration layer. A separate team of software and backend engineers handles the product and infrastructure. Your job is purely AI.

What You'll Build

1. Advanced Retrieval & Indexing Pipeline : Intelligence documents are messy, dense, and massive in volume. Millions of reports spanning decades across HUMINT, SIGINT, IMINT, ELINT, and more. You'll architect a retrieval system that works at this scale: multi-index retrieval across vector, entity, temporal, and graph stores, with intelligent query routing, confidence-weighted fusion, and iterative retrieval loops where the model identifies evidence gaps and fetches additional context.
2. Multi-Agent Reasoning Architecture : Multiple LLMs collaborating on complex intelligence queries: a planning agent that decomposes multi-hop questions, a retrieval agent that pulls evidence from the right indices, a reasoning agent that synthesizes across sources with proper confidence propagation, and a pattern detection agent that mines temporal and relational patterns across millions of events. You'll build the orchestration logic, deciding which model handles what, managing context windows, handling failure modes, and ensuring calibrated, source-attributed answers.
3. Pattern Detection at Scale : Detect patterns no human analyst could find: deployment cycles recurring every 8 months across 2,000 reports, equipment co-location rates across hundreds of instances over years, behavioral baselines and anomalies computed over millions of temporal events. This requires deep thinking about temporal sequence analysis, graph reasoning, and statistical pattern mining, then getting the best possible results from foundation models without fine-tuning.
4. Multi-INT Evidence Fusion : Intelligence comes in wildly different formats, confidence levels, and temporal precisions. You'll build the reasoning layer that fuses evidence across sources, handling contradictory evidence, propagating uncertainty through multi-hop inference chains, and producing answers with proper confidence calibration and chunk-level citations.

What Makes This Hard

1. No fine-tuning. The system runs air-gapped on foundation models only. Every ounce of performance comes from your architecture: query decomposition, retrieval strategy, prompt design, agent orchestration, and result fusion.
2. Unstructured, heterogeneous data. Military intelligence documents don't follow neat schemas. The same entity is referenced differently across sources. Temporal references range from precise timestamps to "sometime in early March."
3. Scale breaks naive approaches. Pattern detection across 1M+ events, graph reasoning over 10K+ nodes, retrieval across 10M+ documents. None of this works with textbook implementations.
4. Confidence matters as much as answers. In intelligence, a wrong answer with high confidence is worse than no answer. Systems must know what they don't know.

What We're Looking For

Non-Negotiables

• You understand LLMs deeply, not as black boxes. You know how attention works, why context window management matters, and how to design prompting strategies that reliably extract structured reasoning from foundation models.
• You can architect multi-agent systems from scratch with proper orchestration, failure handling, and context management.
• Deep expertise in advanced RAG: hypothetical document embeddings, multi-vector retrieval, iterative retrieval loops, re-ranking, confidence-weighted fusion. You know why naive RAG fails and how to fix it.
• First-principles thinker. Given a novel problem, you reason your way to a solution architecture, choosing the right data structures, algorithms, and model interactions.
• Strong Python engineer. Production-quality code, async patterns, maintainable systems.
• Hands-on experience with vector databases, graph databases, inference optimization, and GPU-aware system design.

Strong Differentiators

• Graph neural networks and their application to knowledge graphs and entity reasoning.
• Temporal reasoning, irregular time series, or event sequence analysis.
• Causal reasoning, Bayesian inference, or uncertainty quantification.
• Research publications in NLP, information retrieval, or reasoning systems.
• Familiarity with intelligence, defense, or geospatial domains.

What This Role Is and Isn't

IS:

• Designing AI architecture for a world-class intelligence platform.
• Building multi-agent reasoning systems that push the limits of foundation models.
• Solving retrieval and indexing problems where standard approaches fail.

ISN'T:

• Calling APIs and writing prompt templates.
• Building backends or UI.
• Fine-tuning models on labeled datasets.
• Following tutorials.

Success Metrics

• Automated pattern identification across 90%+ of entities.
• Complex multi-INT query latency under 10 seconds.
• Anomaly detection below 5% false positives.
• System handles 10M+ documents and becomes the primary analyst tool.
• 95%+ reduction in manual analysis time.

Our Team

A small, exceptional group of researchers and engineers building at the frontier of AI for national security. The strongest technical minds we could assemble here because the problem matters and the bar is the highest they’ve encountered.

We’re setting the global standard for AI-driven intelligence analysis by building smarter architectures than anyone else in this space.

"If you’re already thinking about how you’d solve these problems, we should talk!"