EDGE AI POD

What if your printer didn’t just spit out pages, but actually understood them? We walk through a hands-on look at multimodal AI on the edge—how visual-language models read layouts, extract tables, translate content, and reformat documents right where data lives, without shipping sensitive files to the cloud. It’s a practical tour from passive peripherals to active intelligence, with real workflows and measurable speedups.

We share the architecture behind on-device document intelligence: pre-processing that stabilizes inputs, VLMs that localize and reason over text and images, and post-processing that converts outputs into CSVs, charts, and accessibility-friendly layouts. You’ll hear how Qwen 2.5-VL handles complex visual inputs while maintaining strong language performance, and how a Flux-based diffusion setup enables creative generation and targeted edits—from updating dates in greeting cards to changing borders and colors by prompt. Along the way, we unpack quantization with GGUF to run 7B-class models in tight memory, diffusion sampler and scheduler tuning for latency, and NVIDIA-optimized libraries to squeeze more from modest GPUs.

Beyond demos, we dig into business and engineering realities: fine-tuning with enterprise data to reduce hallucinations, building guardrails and fallback paths for reliability, and segmenting large documents to manage VRAM. We also discuss why a companion device—AI PC or smartphone—can orchestrate heavy lifting until printer SOCs catch up, keeping data private and workflows responsive. If you care about document AI, privacy by design, or accessibility features like dynamic type and contrast, this conversation makes the path concrete and actionable.

Enjoy the deep dive? Subscribe, share with a colleague who lives in PDFs, and leave a review with the one edge use case you want us to test next.
Send us Fan Mail
Support the show
Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

What if AI felt like a door you could open, not a wall you had to climb? We dig into how Arduino’s approach—accessibility first, power when you need it—turns the edge AI buzz into a concrete path you can follow, whether you’re a student with a starter kit or an engineer shipping to a fleet.

We walk through a practical four-step journey: try AI through no-code experiments, understand it with pre-trained models, train by fine-tuning or starting from scratch with your data, and build something real that lives beyond a demo. Along the way, we unpack a core principle we call “abstraction without obfuscation”—removing friction while keeping the logic transparent—so you can inspect, modify, and truly own the systems you create. That design philosophy shapes everything from our open hardware portfolio (TinyML-friendly MCUs up to Linux-capable MPUs) to our integrations with popular AI frameworks and community-driven libraries.

You’ll also hear how cloud-native developer tools streamline the messy middle: browser-based workflows, single-device to fleet deployments, secure OTA updates, data collection for predictive insights, and closed-loop model improvement. Plus, we introduce our AI assistant as a coach that explains code, diagnoses bugs, and helps optimize for memory and speed—turning dead ends into learning moments. Real-world validation from a 35-million-strong community and enterprise teams, including automotive innovators, shows how openness and cohesion accelerate the leap from idea to production.

If you care about AI that empowers rather than intimidates, this conversation lays out the playbook. Subscribe, share with a teammate who loves to build, and leave a review telling us the project you’re dreaming about—we might feature it next.
Send us Fan Mail
Support the show
Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

If you’ve ever shipped a model that flew in the cloud and crawled on a device, this conversation is a relief valve. We bring on Andreas from Embedl to unpack why edge AI breaks in the real world—unsupported ops, fragile conversion chains, misleading TOPS—and how to fix the loop with a unified, device-first workflow that gets you from trained model to trustworthy, on-device numbers in minutes.

We start with the realities teams face across automotive, drones, and robotics: tight latency budgets on tiny chips, firmware that lags new ops, and the pain of picking hardware without reliable performance data. Instead of guesswork, Andreas demos Embedl Hub, a web platform and Python library that standardizes compilation, static quantization, and benchmarking, then runs your models on real hardware through integrated device clouds. The result is data you can act on: average on-device latency, estimated peak memory, compute-unit usage, and detailed, layer-wise latency charts that reveal bottlenecks and fallbacks at a glance.

You’ll hear how to assess quantization safely with PSNR (including layer-level drift), why pruning and optimization must be hardware-aware, and how a consistent pipeline across ONNX/TFLite/vendor runtimes tames today’s fragmented toolchains. We also compare Embeddle Hub’s scope to broader end-to-end platforms, touch on non-phone targets available via Qualcomm’s cloud, and talk roadmap: more devices, deeper analytics, and invitations for hardware partners to plug in.

If you care about edge AI benchmarking, hardware-aware optimization, ONNX/TFLite compilation, layer-wise profiling, and choosing devices with data instead of hope, you’ll leave with a practical playbook and a tool you can try today—free during beta. Listen, subscribe, and tell us the next device you want to see in the cloud lab. Your model isn’t done until it runs on real hardware.
Send us Fan Mail
Support the show
Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Clothes that quietly listen to your breath might be the missing link between hospital‑grade vigilance and everyday comfort. We walk through how our team built a textile‑integrated breath sensor that actually works in the wild—embroidered interconnects, 3D‑printed dielectric islands, and a carbonized‑silicon yarn strain gauge stitched into a belt—then taught it to estimate breathing at the edge with TinyML.

We dig into the engineering choices that matter: why flexible interconnects are the “holy grail” for wearables, how a simple peak detector falls apart with drift and burn‑in, and what it takes to turn raw strain signals into reliable features. After screening public datasets that didn’t match our sensor, we built our own: band‑pass filtering in the 0.1–1 Hz range, three‑second windows, normalization, and event‑button labeling for clean ground truth. From there, we used Edge Impulse’s EON Tuner to search architectures and landed on two contenders—a CNN on time‑domain windows and a compact DNN with wavelet features—then deployed both on an STM32L4 with DMA, timers, and CMSIS‑DSP preprocessing.

The results are candid and practical. The CNN was slower but consistently more accurate and robust; the DNN was snappier with lower power but less reliable under offset and noise. Models trained on a different sensor’s data struggled to generalize to our belt, reinforcing a core lesson for smart textiles: sensor‑specific datasets and fine‑tuning are essential. We close by mapping next steps—expanding our dataset, improving transfer across garments and users, exploring hydration prediction, and tightening on‑device optimization—so remote patient monitoring can be seamless, private, and wearable all day.

If you enjoy deep dives into edge AI, embedded systems, and human‑centric health tech, follow the show, share it with a colleague, and leave a quick review to help others find it.
Send us Fan Mail
Support the show
Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

What if the only way to get real gains at the edge is to redesign everything—from the silicon atoms to the app you deploy? That’s the bet Professor-Founder Mohammed Ali made with EMAS, and the results are striking: continuous inference at milliwatts, microsecond wake/sleep cycles, and real benchmarks that hold up against the best in class while burning a fraction of the energy.

We walk through how a RISC-V core, dual AI accelerators, and an MRAM/RRAM-backed memory system work together to keep weights on-chip, slash data movement, and power-gate aggressively without losing state. The compiler handles pruning, quantization, and on-the-fly compression to achieve around 1.3 bits per weight without torpedoing accuracy, while a custom memory controller mitigates non-volatile quirks like endurance and read variability. Instead of chasing TOPS, the stack optimizes bandwidth, dataflow, and timing to match the realities of sensors and batteries.

The story gets especially interesting with drones. Since propellers—not processors—dominate energy use, EMAS applies tiny AI to the control problem, redistributing load across rotors in real time and extending flight endurance by 60% or more in hardware-in-the-loop simulations. We also dig into wearables and time-series workloads like ECG, audio, and vibration, where sparse sampling pairs perfectly with microsecond power gating. If you build at the edge, the dev experience matters: you’ll hear about the virtual dev kit with remote access to real silicon, a compact evaluation board with modular sensors, and an SDK that plugs into TensorFlow, PyTorch, and Zephyr. Advanced users can map trained models via a CLI; newcomers can lean on a NAS-based flow that proposes architectures meeting strict memory and power budgets.

If you care about edge AI, battery life, and shipping reliable products, this conversation is a blueprint for co-designing across the stack to unlock 10–200x energy gains without giving up performance. Subscribe, share with a teammate who owns your edge roadmap, and leave a review with the one use case you’d optimize first.
Send us Fan Mail
Support the show
Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org