(Senior) Software Engineer, AI/ML (m/f/d)

Location: Berlin/remote
Working Hours: Full-time/Part-time
Start Date: Immediately or by arrangement

About Us

Project Q GmbH (“Q”) is a Germany-based DefenceTech company with offices in Berlin and Munich. We network market-available sensors into a tactical IoT backbone that reliably provides armed forces and security agencies with a cross-domain information superiority—automated, scalable, and cost-efficient.

Our solutions integrate seamlessly into existing defence and civil infrastructures, offering fast data fusion and AI-enhanced analytics to deliver robust situational awareness in dynamic environments. We build platform-agnostic systems that enable real interoperability across domains.

As a Software Engineer in AI & ML, you’ll own the models that turn raw sensor and language data into actionable detections, classifications, and summaries that feed the common operating picture. You’ll work across modalities, taking models from training pipeline through to edge inference. The models have to be right, the inference has to run in real time on constrained hardware and scale for higher-compute environments, and the output has to hold up under unexpected field conditions (domain shift, sensor degradation, jamming, adversarial inputs).

What you’ll do

Model development and training

• Design, train, and evaluate models for the task at hand—detection, segmentation, classification, captioning, VQA, ASR, or signal classification—choosing architecture and training strategy based on the data and the deployment target, not the latest leaderboard.
• Build the data side as a first-class engineering problem: curation, labeling workflows, augmentation, synthetic data generation, and hard-negative mining.
• Own model evaluation: task-appropriate metrics, slice-based analysis, OOD and adversarial test sets, and red-teaming for failure modes that matter operationally.
• Apply fine-tuning, distillation, quantization, LoRA/adapters, and prompt engineering pragmatically to get the right behavior into the deployable footprint.

Multi-modal integration

• Fuse signals across modalities (EO/IR imagery, acoustic, radar returns, RF, text/voice reports) using late-fusion ensembles, cross-attention architectures, or VLM/foundation-model backbones.
• Build cueing and grounding logic so models cooperate with downstream tracking and fusion (e.g. VQA-style queries against thermal feeds, language-grounded retrieval over historical detections, acoustic classifications conditioning a vision model).
• Handle the practical glue: synchronization of multi-rate streams, frame and coordinate consistency, and confidence calibration across heterogeneous models so downstream consumers can actually trust the numbers.

Real-time systems and deployment

• Stand up inference services as part of a real-time streaming system (gRPC, TCP, message buses): batching, backpressure, out-of-order handling, model versioning, graceful degradation.
• Profile and optimize to hit latency and power budgets on gateway and edge hardware (Jetson, embedded GPU, NPUs).
• Run simulation, replay, and hardware-in-the-loop tests; own the gap between model-in-a-notebook and model-in-the-field.
• Monitor deployed models for drift, degradation, and silent failure; build the loops that get field data back into training.

Collaboration

• Partner with sensor fusion, software, and field engineering teams.
• Contribute to ML standards and infrastructure that scale across the product portfolio, and to interoperability with external systems aligned to NATO standards.
• Mentor teammates on modern ML practice and the pragmatic side of taking models to production.

You should apply if you

• Have shipped ML/AI systems into real operational environments, not just notebooks or benchmark runs.
• Have deep, hands-on expertise in at least one of: computer vision, current-generation language models, visual question answering / VLMs, or acoustic/seismic/remote-sensing signal processing (and can defend your modeling, data, and training choices for a given problem).
• Are fluent in Python, ML frameworks, and are comfortable building real-time inference components in applications, not just training scripts.
• Can diagnose why a deployed model misclassifies, hallucinates, or degrades, and fix the root cause.
• Communicate clearly with sensor fusion, software, and forward-deployed engineers alike; you’re a translator as much as a builder.

You’d be a great fit with

• Experience with foundation models, VLMs and SOTA for-use focused models (CLIP, SAM, LLaVA-family, Qwen-VL, YOLO, etc.) and the practicalities of fine-tuning and serving them.
• Background in maritime surveillance, counter-UAS, GNSS-denied navigation, or ISR.
• Experience with self-supervised or contrastive pretraining, domain adaptation, continual learning, or active learning on operational data.
• Familiarity with ROS 2 and real-time middleware.
• Exposure to radar signal processing, acoustic beamforming, or RF/spectrogram-based classification.
• Experience deploying to constrained edge targets (NVIDIA Jetson, embedded Linux, microcontrollers).
• Familiarity with defense or safety-critical operational environments.

We Offer

Be part of our growth story and help build a leading defence technology company tackling complex technological challenges, contributing to European sovereignty and security.

• Opportunities for professional and personal development
• Flexible working hours and remote work options
• Competitive salary and additional benefits

Have we sparked your interest?

We look forward to receiving your CV, including your earliest possible start date, at:

HR@project-q.ai

We can’t wait to meet you!