Multi-modal perception, 3D reconstruction, and generative deep learning — from IEEE publications to production deployment.

I build AI systems that fuse multi-sensor data (LiDAR, radar, cameras, GPS, IMUs) into robust 3D understanding—optimized for production deployment under real-time constraints.

My work spans production-scale perception systems, 3D reconstruction pipelines, and generative modeling with diffusion and flow-matching techniques. I bridge classical computer vision with modern deep learning, working across object detection, segmentation, vision-language models, and multi-modal fusion.

Published researcher in IEEE journals on diffusion models and state space models. Technical thought leader with 24+ articles on foundation models, attention optimization, and deep learning (read by thousands of engineers). BEO Expert (Bosch Expert Organization) providing cross-divisional AI consulting.

Before specializing in AI, I built FPGA-based test instrumentation and designed radar electronics—a hardware foundation that shapes how I think about latency, memory bandwidth, and real-world constraints.

I don't just run models; I architect them. I optimize foundation models for production deployment, adapt them for specific sensor configurations, and squeeze performance through quantization, profiling, and algorithmic refinements. I lead teams, make architecture decisions, and translate cutting-edge research into systems that ship.

• Published research implementations: SambaMixer (Mamba state space models for battery health), RGB-D-Fusion (diffusion models for depth estimation), VoloGAN (GAN-based domain adaptation)
• Deep learning framework: DeepSaki—my TensorFlow/Keras toolkit for rapid prototyping and experimentation with custom layers, losses, and architectures
• Dev environment automation: dotfiles (Neovim, tmux, shell configs) and linux-forgeup (lightweight system bootstrap tool)
• ML experiments & notebooks: explorations of generative models, foundation model fine-tuning, and computer vision techniques
• Personal website & portfolio: Jekyll-based static site showcasing projects and technical background

Most production work happens behind closed doors—but you can explore my thinking here:

• 📄 LinkedIn — technical leadership, BEO Expert consulting, IEEE-HKN presidency
• 📝 Medium @SaschaKirch — 24+ technical deep-dives (FlashAttention, Mamba, foundation models)
• 🌐 Portfolio — project showcase, publications, professional background

If you care about squeezing every drop of performance from models and turning multi-sensor fusion into production reality, let's connect.