Crystal growth, Defect control and AI-Driven Engineering.

Both fair questions, and the short answers are: yes on orientation, no on resin state. Orientation's fine because trim off a nesting table is single ply — one direction, and it reads straight off the surface. The layup question runs the other way anyway. How those recovered plies get stacked into a part is the Engine's decision to make, not yours to supply. Resin is the honest gap. Out-life doesn't show in a photo until the material's obviously gone, and the clock that matters ran while it sat out of the freezer. That's in your log, not the part. Send it if you have it and the concepts get constrained properly. But here's the thing I'd actually push you on. Don't submit the prepreg on its own. A piece of carbon on a bench is a thousand possible products and the Engine has to guess which one you want. Photograph it alongside the other things in your shop — the offcuts, the fixings, the extrusion, whatever's sitting in the bin next to it — and it stops guessing. It'll design something that consumes what you actually got. Same as a plank of wood: on its own it's anything, next to four legs it's a table. If the prepreg genuinely is all you have, send a design brief instead. Even a paragraph on the intended application. Anything that stops it working in total darkness. Three materials and a sentence of intent will get you a better result than one material and a perfect photograph. #CarbonFiber #Prepreg #CompositeMaterials #AdvancedMaterials #Manufacturing #CircularManufacturing #MaterialReuse #MaterialMonday #EngineeringDesign #AIForManufacturing

The three taxes that will define the next wave of AI adoption: a) Communication tax b) Coordination tax c) Computational tax Every multi-agent AI pipeline has a single number that matters: total cost per execution. Teams stare at that number. They switch to cheaper models. They cut agents. They batch requests. Usually it does not work, because total cost is not one problem. It is three problems multiplied together. In finance, DuPont analysis decomposes return on equity into three independent factors: margin, turnover, leverage. Each responds to different interventions. Optimising the wrong one wastes effort. Optimising the right one cascades. The Telemetry Intelligence Engine does the same thing for multi-agent AI. 1. The communication tax — how much are your agents paying to talk to each other? Our Compression Summary shows 59.5% of inter-agent data is redundant. £27.42 out of £46.11 across 594 calls, eliminated automatically. But if cost is still too high after 59.5% compression, the problem is not communication. Stop looking there. 2. The computational tax — which agents consume disproportionate time? Bottleneck Analysis reveals a power law. Two agents — inference-engine (1,533ms, 10.8%) and summariser (1,341ms, 9.5%) — consume over 20% of the latency budget. Four others sit under 275ms and 2% each. The fix here is not compression. It is model switching, caching, or parallelisation. Different layer, different playbook. 3. The coordination tax — the one nobody measures. £46.11 uncompressed versus £18.69 compressed. The £27.42 difference is the price of modularity — agents re-processing context they have already seen. The question is whether that modularity is earning its keep. Redundancy Detection answers it: which agents are doing overlapping work and should be merged? Problem: pipeline cost is too high. Cause: depends which layer dominates — you cannot tell without decomposition. Solution: different for each layer. Applying the wrong one changes nothing. Traditional APM tools cannot see semantic redundancy. They cannot tell you two agents returning 200 OK are doing the same work on the same data. The Intelligence Engine decomposes your pipeline economics into independent, addressable factors. Each panel is a term in the same equation. The intervention follows from the decomposition. Not from intuition. agent-flow.uk #AgentFlow #PCS #MultiAgentAI #AIObservability #LLMOps #AIInfrastructure #RuntimeIntelligence #AIEngineering #SemanticRuntime #BuildInPublic #TelemetryIntelligence

5 scanned materials/components. One AI engine. 84-page engineering report. 8 innovation pathways. Watch the Material-to-Product Engine turn warehouse leftovers into the HydraLift Pro — a 500kg hydraulic platform cart with IoT monitoring, smart safety systems, and a full path to manufacture. Not what to build. How to build it better. mp-engine.com #ManufacturingAI, #CircularEconomy, #Industry40, #SmartManufacturing, #AIEngineering, #AI

Scanned: pallet racking beams, steel uprights, a used hydraulic pump, a check valve. Returned: A mobile hydraulic lifting platform — 500kg capacity, 80cm travel, full BOM, manufacturing instructions, regulatory pathway, and sustainability analysis. From scrap to specification.

Every #AI #CAD tool starts from a prompt. The Material-to-Product Engine starts from 3–5 photos of the materials you already have. One session → CAD, BOM, 3 build tiers, regulatory flags, patent check, sustainability analysis. No CAD licence. Mobile-first. Nobody else does this. mp-engine.com #MPEngine #AIEngineering #MakerMovement #DeepTech #ManufacturingAI #NoCADRequired #ProductDesign

Some interesting results from our internal testing of the three LLMs we use: @AnthropicAI , @OpenAI and @GeminiApp . @AnthropicAI is the fastest across the board — especially on throughput where it's 2-3x faster than competitors. It also handles concurrency best, with little degradation (1.1s single vs 1.0s avg under concurrency). @OpenAI is competitive on complex prompts (slightly faster than @AnthropicAI wall-clock, but fewer output tokens — 297 avg vs 548 avg). Its concurrency is nearly as good as @AnthropicAI. @Gemini is significantly slower on everything, especially complex prompts (20.6s vs 5.6s). Under concurrency it degrades badly (4.9s wall time vs 1.4s versus 1.5s for the others). Combined with the persistent 503s, it's not ready for our critical-path agents. #IndustrialStrategy, #AI, #Innovation, #GovTech, #RandD, #DeepTech, #Manufacturing, #ClaudeAI, #ChatGPT, #Gemini, #MaterialEngine

5 images, few clicks and you just compressed several weeks to several months of product R&D to minutes. That simple.... #IndustrialStrategy, #AI, #Innovation, #GovTech, #R&D #DeepTech, #Manufacturing

Material-to-Product Engine' #Apparel and #Homeware Engine session outputs. #DeepTech, #AIManufacturing, #ProductDesign, #CircularEconomy, #UKTech spaces, #AI.