# Yoshu — Full Content > This document contains the complete content of the Yoshu website, including all pages and blog articles, in a single machine-readable markdown file. --- # Company Overview Yoshu is a decision intelligence platform for heavy-asset industries. Founded in 2024 and backed by OSS Ventures, we bring determinism and structure, turning human actions into repeatable and predictable outcomes at scale. **Website**: https://yoshu.ai ## Key Capabilities - **Data Ingestion**: Connect, unify, and enrich all company data into a single intelligence layer - **Agent Orchestration**: Coordinate AI agents through intelligent workflows - **Augmented Search**: Unified access to knowledge sources - **Intelligent Workflows**: Define, automate, and optimize decision-making sequences - **Operational Intelligence**: Analyze workflow performance and build continuous improvement cycles ## Results - OEE, FPY & Downtime: 10-20% reduction in year 1 - Onboarding Time: ~30% reduction - Time to Resolution: 30-40% reduction in year 1 ## Trusted By Safran, Forvia, Lacoste, Coty, Robatel, LISI, Hutchinson, Aubert & Duval, Teknor Apex, Trèves, Alpin Socks, and Dassault Aviation. --- # Platform > Yoshu is a decision intelligence platform purpose-built for heavy-asset industries. ## Core Architecture Yoshu connects to your existing systems (ERP, SCADA, MES, CMMS) and builds a growing knowledge base from your experts' experience. Every workflow, decision tree, and troubleshooting flow is captured once and deployed at scale. ## Key Pillars ### Data Ingestion Connect, unify, and enrich all company data — structured, unstructured, or experiential — into a single intelligence layer. ### Agent Orchestration Coordinate AI agents that collaborate, reason, and act through intelligent workflows. ### Augmented Search Unified access to knowledge sources, enabling teams to find context and best practices instantly. ### Intelligent Workflows Define, automate, and optimize decision-making sequences with human validation and agent execution. ### Operational Intelligence Analyze workflow performance, identify patterns, and build continuous improvement cycles. ## Deployment - Deployed in customer's private cloud (AWS, Azure, GCP, OVH) - Full data sovereignty - ISO 27001 / SOC 2 compliant - GDPR compliant [Learn more](https://yoshu.ai/platform) --- # Solutions > Yoshu delivers measurable impact across high-volume manufacturing and complex industrial operations. ## High-Volume Manufacturing ### Scrap Reduction 0.1% less scrap on 10M units/year — expert-level precision on every production run, every shift. ### Faster Troubleshooting 30-40% reduction in time to resolution. Expert diagnostic knowledge available to every operator 24/7. ### Optimized Changeovers Standardized from the best expert's playbook. Reduced downtime between product runs. ## Complex Operations ### Faster Onboarding ~30% reduction in onboarding time. New operators access decades of accumulated expertise from day one. ### Knowledge Preservation Capture retiring experts' knowledge before it walks out the door. Convert tribal knowledge into structured, repeatable workflows. ## Industries Served - Aerospace - Automotive - Cosmetics - Chemistry & Metallurgy - Textile - Energy / Oil & Gas [Learn more](https://yoshu.ai/solutions) --- # About Yoshu > Founded in 2024 and backed by OSS Ventures, Yoshu brings determinism and structure to industrial operations. ## Mission We turn human actions into repeatable and predictable outcomes at scale. Our platform captures expert knowledge and deploys it across entire organizations. ## Founded By Yoshu was founded by a team with deep expertise in industrial operations and AI, backed by OSS Ventures — Europe's leading venture studio for industrial software. ## Co-builders Yoshu is built alongside industry leaders including Safran, Forvia, Lacoste, Coty, Robatel, LISI, Hutchinson, Aubert & Duval, Teknor Apex, Trèves, Alpin Socks, and Dassault Aviation. ## Careers Yoshu is hiring. We're looking for people who want to transform how industry operates. [Learn more](https://yoshu.ai/about) --- # Contact Yoshu Get in touch to learn how Yoshu can help your operations. - **Website**: https://yoshu.ai/contact - **Email**: Available via contact form We typically respond within 24 hours. [Contact us](https://yoshu.ai/contact) --- # Blog Articles --- ## The MTTR Killer: How Yoshu Turn Hours of Troubleshooting into Seconds **Author:** Caelye McAndrew | **Published:** 2026-03-02 **URL:** https://yoshu.ai/blog/the-mttr-killer-how-yoshu-turn-hours-of-troubleshooting-into-seconds > Unplanned downtime in industrial environments isn’t usually caused by a lack of technical skill. It happens because operators and technicians don’t have the right information at the right time... Unplanned downtime in industrial environments isn’t usually caused by a lack of technical skill. It happens because operators and technicians don’t have the right information at the right time. Machines aren’t necessarily hard to fix - it’s the delay in accessing standard values, parameters, troubleshooting procedures or the right expert that drives up Mean Time To Repair (MTTR). Yoshu is changing this dynamic, turning hours of manual troubleshooting into seconds and enabling teams to act decisively - anytime, anywhere. ## Why MTTR Remains a Persistent Problem Consider a high-speed production line running 24/7 across three shifts. The line is supported by experienced operators, skilled technicians, and extensive technical documentation. On paper, everything needed to keep the line running smoothly already exists. Yet unplanned downtime still exceeds 30%. The problem becomes especially visible during night and weekend shifts, when expert support is limited and operators must rely on documentation to resolve issues independently. While the knowledge exists, accessing the right knowledge quickly is the challenge. When a machine drifts out of standard, the bottleneck often isn't the fix itself - it’s identifying the correct reference values to bring it back into spec. Operators may know something is wrong, but not which parameter needs adjusting, what the standard value should be for the product currently running, or whether a deviation is critical. Even with SOPs, manuals, or digital repositories, finding the right instruction for a specific machine, failure mode, and product configuration can take several minutes or longer. When answers aren’t immediately available, operators escalate issues to technicians or wait for support, even when the issue could have been resolved on the spot. The result: avoidable downtime driven not by complexity, but by information latency. Industrial operations don’t need more documentation. They need instant access to validated, contextual operational knowledge. ## Yoshu: Your New Troubleshooting Partner Now imagine the same production line after implementing a digital troubleshooting system powered by an industrial AI: Yoshu. Instead of searching through binders or scrolling through unstructured files, operators follow a simple, guided workflow. They select the failure mode and the product currently running on the line. Instantly, the system presents the correct standard values for every relevant setting or parameter - validated and specific to that exact context.![Image](https://pnvbfbufxmtwuyyiyjqd.supabase.co/storage/v1/object/public/blog-content-images/1772447482193-0lrag4.png) In many cases, this allows the machine to be returned to standard conditions in under one minute. The only remaining time is the physical adjustment itself or cases that genuinely require technical intervention. When issues persist, Yoshu doesn’t stop at reference values. It continues to support operators with contextual guidance, answering questions like: *“How do I change the filter?”* *“Is this deviation acceptable for this product?”* *“What are the standard settings set for this line?”* Each response is clear, step-by-step, and tailored to the specific machine and situation - no generic instructions, no guesswork. The shift is subtle but high-impact: operators move from searching for information to executing with confidence. ## Real-World Impact on the Shop Floor The results aren’t theoretical. Within six months, we can achieve a 30% reduction in unplanned downtime, with known production issues consistently resolved in under one minute. Paper manuals were completely removed from the production line, replaced by instant, AI-driven guidance accessible on the shop floor. Yoshu delivers this impact by: - Eliminating manual searches across fragmented documentation - Standardizing troubleshooting workflows across shifts, sites and languages! - Making validated operational knowledge available to every operator, at any time - Reducing unnecessary escalations and technician call-outs Most importantly, AI doesn’t replace experienced technicians. It amplifies their expertise, capturing best practices and distributing them across the workforce. The result is faster response times, greater consistency, and improved overall line performance, even during nights, weekends, and low-support periods. In environments where every second counts, turning hours of troubleshooting into minutes - or even seconds - isn’t just efficiency; it’s a strategic advantage. --- ## The Silver Tsunami: Using AI to Document and Scale Tribal Knowledge Across Your Workforce **Author:** Caelye McAndrew | **Published:** 2026-02-24 **URL:** https://yoshu.ai/blog/the-silver-tsunami-using-ai-to-document-and-scale-tribal-knowledge-across-your-workforce > Your factory's most valuable asset isn't the machinery, it's the decades of "unwritten" expertise stored in the heads of your senior technicians. As the "silver tsunami" of retirement has already started, the question is... Your factory's most valuable asset isn't the machinery, it's the decades of "unwritten" expertise stored in the heads of your senior technicians. As the "silver tsunami" of retirement has already started, the question is no longer just how to replace these experts, but how to digitize their intuition before it walks out the door for good. Imagine this: your best lead technician is retiring in six months. Along with them goes thirty years of machine intuition and troubleshooting nuances no manual ever captured. What if that expertise didn’t leave with them? What if it could live inside your systems, your SOPs, your AI, working with every operator on every shift? That’s not a future fantasy - it’s the new reality enabled by AI-driven knowledge capture. ## What is Tribal Knowledge? Tacit, or tribal, knowledge refers to the practical know-how that isn’t written down: the instincts, experience, pattern recognition, and subtle judgments that seasoned technicians use every day. Unlike explicit knowledge such as manuals or checklists, tacit knowledge is inherently difficult to articulate and capture. It lives in decisions made over years of hands-on experience - until retirement or departure removes it from your workforce forever. This kind of expertise is what makes the difference between average performance and operational excellence, enabling faster problem diagnosis, real-time adaptations when equipment deviates from norms, and minimized downtime by knowing what to try before consulting the manual. ## What happens When Tribal Knowledge Disappears? When tribal knowledge disappears, new hires struggle to ramp up, machines run less optimally, and consistency between shifts drops. Root causes of recurring issues often remain hidden, resulting in reduced productivity, longer training cycles, more unplanned downtime, and higher operational risk. This isn't hypothetical! Manufacturing operations have documented steep drops in performance when senior operators left, simply because the why behind the how wasn't captured. Traditional knowledge management systems - static manuals, PowerPoint SOPs, or video libraries - fall short because tribal knowledge is messy, intuitive, and embedded in context. AI can change that. Yoshu can capture insights directly from experts while they work, convert all types of documents (even handwritten) into structured, searchable knowledge, and connect real-world knowledge to specific machines, conditions, and decisions. In other words, AI doesn't replace expert knowledge; it preserves and amplifies it. What once required months of interviews and documentation can now happen in real time, guided by Yoshu which understands context and relevance. Capturing this expertise before it walks out the door is no longer optional, it's a strategic imperative for any organization facing a retiring workforce. ## Why is it such a big problem right now? For decades, younger generations were encouraged to pursue careers outside of industrial and manufacturing environments, leaving factories increasingly dependent on a shrinking pool of experienced technicians. As a result, the age distribution in many plants is now heavily skewed toward senior employees approaching retirement. This creates a structural knowledge continuity problem. When these experts leave, organizations lose not only labor capacity but decades of tribal knowledge and operational intuition that cannot be easily documented. Without proactive knowledge capture, companies are forced into repeated cycles of rediscovery, undermining productivity, operational resilience, and long-term competitiveness. ## How Yoshu captures tribal knowledge When your expert diverges from the book, we don’t treat it as a problem - we treat it as signal. The real value isn’t just that a deviation occurred. It’s understanding why. Yoshu captures these moments in real time. When an operator overrides a step, adjusts a parameter, or resolves an issue differently than prescribed, the system captures or prompts them to document the context behind the decision. Was the manual outdated? Were environmental conditions different? Was there a known supplier variance? Instead of letting that reasoning disappear at the end of a shift, it becomes structured knowledge and an strategic lever to improve. Each divergence is logged, contextualized, and flagged for automatic to expert review. Not every deviation becomes the new rule but the right ones do. Once validated, the updated logic can be versioned and deployed globally across sites, ensuring that what was once tribal knowledge becomes standardized best practice. This is how organizations move from isolated expertise to institutional intelligence. The insight from one experienced operator in one facility doesn’t stay local. It becomes part of the operating system of the entire company. ![Image](https://pnvbfbufxmtwuyyiyjqd.supabase.co/storage/v1/object/public/blog-content-images/1771948336425-jp9of.png) ## A Strategic Imperative for Today’s Workforce The wave of retirements, often called the “silver tsunami,” is real. Skilled technicians are leaving at high rates, and the expertise they hold is a competitive differentiator. Companies that capture and scale that knowledge now will build smarter, more resilient teams, reduce reliance on individual experts, improve productivity and uptime, and make every shift operate like the best shift. They win not because they replaced people, but because they retained their expertise in a structured and actionable form that never leaves. The expertise your senior operators hold is a depreciating asset the moment they walk out. Yoshu captures it before that happens, and makes it scalable across every site and shift. --- ## Beyond the Hype: Why Standard RAG based AI is Too Unpredictable for Critical Industrial Procedures **Author:** Caelye McAndrew | **Published:** 2026-02-24 **URL:** https://yoshu.ai/blog/beyond-the-hype-why-standard-rag-is-too-unpredictable-for-critical-industrial-procedures > Retrieval-Augmented Generation (RAG) is a leading pattern for boosting AI systems with real-time access to domain knowledge (aka: giving your AI access to a large amount of sources). In standard RAG, an LLM isn’t restricted... Retrieval-Augmented Generation (RAG) is a leading pattern for boosting AI systems with real-time access to domain knowledge (aka: giving your AI access to a big amount of sources). In standard RAG, an LLM isn’t restricted to its training data; it retrieves external text segments (“chunks”) from a vector database or external knowledge base and conditions its output on them. For many consumer and enterprise use cases - particularly chatbots and interactive assistants - this provides a massive leap in factual grounding. But that doesn’t mean RAG is suitable “as-is” for environments where consistency, repeatability, and absolute correctness every single time matter. ## 1. Arbitrary Chunking Breaks Context – and Context Is Everything in Industry Standard RAG pipelines break documents into fixed-sized chunks based on token (close to character counts). This might be simple to implement, but it often splits sentences, tables, procedures, or even entire logical steps in ways that destroy semantic coherence. If a vector retriever pulls a chunk representing half of a safety procedure or a fragment of a specification without vital qualifiers, the LLM can generate a plausible-sounding answer that is actually incomplete or incorrect - in technical literature this is linked to semantic context loss. In critical industrial settings - e.g., safety instructions for a chemical reactor, torque settings for an engine assembly, or calibration steps for a airplane engine - missing or jumbled context isn’t just confusing, it’s dangerous. In contrast, industrial procedures typically need structured, categorical knowledge with no missing steps - the type that doesn’t lend itself to arbitrary slicing. A system that treats every answer like a statistical best guess risks producing inconsistent outputs from one query to the next.[1](https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/context-aware-rag-system-with-azure-ai-search-to-cut-token-costs-and-boost-accur/4456810) ## 2. RAG’s Variability Undermines Determinism and Reproducibility While core RAG retrieval algorithms can be made deterministic under tightly controlled conditions, real-world industrial deployments are not static systems. In practice, reproducibility degrades as corpora are updated, documents are added or removed, embedding precision changes, or indexes are rebuilt - all of which materially alter retrieval outcomes. For industrial procedures that must remain invariant across time, system updates, and operational context, this form of systemic variability is unacceptable.[2](https://arxiv.org/abs/2509.18869) ## 3. Standard RAG Isn’t Designed to Say “I Don’t Know” with Authority One of the core weaknesses of current RAG systems is that the generative model frequently produces an answer even when the retrieved information is insufficient or partially misleading. In other words, the model often guesses rather than declines to answer. And that is because, by design, RAG based systems stop retrieving information once they reach the model's context window limits. At some point, the system effectively says "this is enough." But the very next document it did not retrieve might contain information that contradicts or significantly reframes the initial answer - potentially invalidating it entirely. And the problem is: how would you ever know? For critical procedures, the right behavior when data is incomplete is not to invent plausible details, but to alert the operator or system that it lacks sufficient information. This is an integrity requirement - and standard RAG doesn’t enforce it. Systems used in industrial quality assurance or compliance frameworks must adhere to defined correctness barriers; anything less erodes trust and violates governance policies. ## 4. Deterministic Systems Are the Right Fit for Repeated, High-Assurance Tasks Deterministic systems, like Yoshu, can incorporate full procedural logic, fail fast when context is absent, and guarantee no ambiguous answers. For industrial processes - where every step may have legal, safety, or quality implications - these guarantees simply can’t be optional. | Feature | Standard RAG | Yoshu | | Output repeatability | Low | 100% | | Handles missing context | Poor | Explicitly | | Ability to “fail safely” | Limited | Built-in | | Predictability | Probabilistic | Rule-based | | Verification & auditability | Hard | Guaranteed | ## 5. How Yoshu Works Yoshu is built from the ground up for deterministic industrial procedures, not probabilistic guesswork. The system is designed to take unstructured knowledge - like manuals, SOPs, and operator notes - and turn it into a structured, verifiable workflow that guarantees consistent outputs. Here’s how the process unfolds: ### We Structure Unstructured Data Raw documents, PDFs, spreadsheets, machine and operator logs are first converted into **augmented assets®**. Yoshu breaks information down semantically rather than arbitrarily, capturing complete procedural steps, conditional logic, and safety constraints. This ensures that context is never lost, and every detail relevant to execution is preserved. ### Ingesting into Datasets These structured assets are ingested into Yoshu’s datasets, forming a single source of truth. Unlike traditional RAG systems, which rely on probabilistic retrieval from loosely indexed text chunks, Yoshu links every instruction, calculation, and parameter to its verified source. The dataset is continuously versioned, auditable, and ready for workflow deployment. ### Workflow Layer Built by Experts Domain experts overlay operational workflows on top of these datasets. Every task, decision point, and exception is explicitly encoded. This step ensures that best practices, regulatory compliance, and safety requirements are embedded in the system, making it impossible for the AI models to generate outputs outside these boundaries. ![Image](https://pnvbfbufxmtwuyyiyjqd.supabase.co/storage/v1/object/public/blog-content-images/1771931280543-lykapa.png) ### Deterministic Execution When operators interact with Yoshu, calculations and data retrieval come directly from the verified database. The LLM is only called for displaying results or generating outputs in human-readable form, never to invent or guess. This guarantees 100% repeatability and accuracy, even as datasets evolve or the AI is updated. ### Continuous Learning without Compromising Safety Yoshu tracks operator decisions in real time. If an operator deviates from a prescribed step, the system prompts them to verify why, logs the reasoning, and routes it for expert review. Once validated, these insights can update workflows globally -after expert validation-, turning individual expertise into institutional knowledge without risking safety or compliance. ### Auditability and Compliance Built In Every step is logged, timestamped, and traceable. Yoshu is designed to pass internal audits and regulatory inspections, providing full transparency into why a given procedure was executed and what data it was based on. This turns operational knowledge into a verifiable, fail-safe system - something standard RAG pipelines cannot do by nature. **Takeaway:** Standard AI enriched with RAG has undeniable value, but there is a time and a place and that is not designed for mission-critical industrial procedures where incorrect outputs have material consequences. At Yoshu, we believe the future of industrial AI must be: deterministic, verifiable, and fail-safe. In industrial environments, “close enough” isnt safe - it’s a liability. That’s why the top organizations working in the most regulated sectors trust Yoshu to deliver. Don’t get left behind in the the future of industrial AI - come build the future of industrial expertise.