Who is Castolin Eutectic?
Castolin Eutectic is a global industrial leader founded in 1906 — over a century of expertise in maintenance, surface protection, and special joining through welding, brazing, and coating technologies. Headquartered in Kriftel, Germany, the company serves industries ranging from mining and cement to aerospace and power generation, with a worldwide sales force of 300+ representatives and a product portfolio exceeding 20,000 items.
That depth of experience is a genuine competitive advantage. But it also creates a formidable internal challenge: the knowledge that makes Castolin Eutectic indispensable to its customers lives in thousands of documents, past project reports, technical data sheets, and the minds of a relatively small group of subject matter experts — not in the hands of every rep who needs it, every day.
The Situation: Knowledge at Rest
Castolin Eutectic’s sales teams operate in technically demanding environments. A rep engaging with a cement plant, a steel mill, or a hydropower operator needs to speak confidently about alloy selection, application methods, metallurgical trade-offs, and proven precedents from similar projects. That conversation can take days to prepare well — or it can go sideways if the rep is working outside their area of expertise.
The knowledge to handle these situations existed inside the organization. But it wasn’t accessible in any structured way. Finding the right data sheet, the right past project reference, or the right application note meant relying on local experts, manual searches across disconnected systems, or simply not finding it at all.
Three specific problems surfaced repeatedly:
Product and expertise complexity. Castolin Eutectic’s portfolio spans hundreds of products across welding, brazing, and wear protection — each with deep technical requirements. Reps are expected to know not just what a product does, but when to recommend it, how it behaves in specific applications, and what the right parameters are. The sheer breadth made consistent competence across the team nearly impossible.
Global knowledge silos. With teams distributed across dozens of countries, the organization had accumulated enormous collective expertise — but no reliable way to surface it. Best practices developed in one region rarely reached another. That knowledge lived in people’s heads or scattered documents, not somewhere accessible.
Time-intensive proposal preparation. Building a proposal required reviewing drawings, cross-referencing past projects, checking product specs, and coordinating with internal experts — a process that could consume days and still produce inconsistent output across the team.
The Challenge: Making Expertise Accessible Without Losing Control
The underlying problem wasn’t that Castolin Eutectic lacked knowledge. It was that the knowledge was locked — distributed across too many formats, systems, and individuals to be reliably accessible at the point of need.
Any solution had to meet a non-trivial bar. The content was highly technical, spanning metallurgy, application engineering, and industry-specific best practices. Retrieval had to be precise — not just surface content, but surface the right content with citations, so reps could verify before they acted. And the entire system had to run inside Castolin Eutectic’s own private cloud tenant, respecting strict enterprise security and compliance requirements.
Standard process improvements — better folder structures, more documentation, clearer naming conventions — had already been tried. They helped at the margins but couldn’t deliver on-demand answers synthesized from multiple internal sources in real time. That required a different kind of system.
The Solution: SalesBuddy
We partnered with Castolin Eutectic to design and build SalesBuddy — a Retrieval-Augmented Generation (RAG) system that gives every rep access to the full depth of the organization’s knowledge, from any device, in natural language, in seconds.
Scalable Data Pipeline: Extraction, Structuring, and Enrichment
The first challenge was getting decades of institutional knowledge into a form a retrieval system could actually use. We built a pipeline to ingest content from multiple internal sources — the Solutions Portal, technical data sheets, training materials, Confluence, SharePoint, and historical project documentation. Raw documents were extracted, structured, and enriched through a semi-supervised chunking process: automated ingestion was combined with manual chunk curation in high-value areas to ensure that the most technically nuanced content was represented with the precision the downstream retrieval system required. Key decisions — like how to segment complex metallurgical procedures or cross-reference application examples — were made deliberately, not left to defaults. The result was a knowledge base that accurately reflected how Castolin Eutectic's experts actually think about and explain their domain.
Search & Retrieval Optimization: The Hardest Problem
Retrieval quality was the make-or-break challenge. A system that returns plausible-sounding but imprecise answers in a technically demanding sales context is worse than no system at all — it erodes trust and creates liability. We invested heavily here: combining semantic vector search with keyword-based retrieval (BM25) and tuning the weighting between them across different query types. Manual chunk boosts were added for high-priority content to ensure that foundational knowledge surfaced consistently. Critically, we built a comprehensive evaluation suite using Langfuse — instrumenting every retrieval step to track answer relevance, citation accuracy, and model behavior on complex multi-document queries. This gave us a systematic, data-driven feedback loop to identify and close retrieval gaps over time, rather than relying on qualitative impressions. The evaluation framework became a core part of the ongoing product — not a one-time check.
Chat Interface: From Query to Actionable Answer
The front-end experience was designed around a single workflow constraint: a rep needs to go from question to verified, actionable answer in minutes, not hours. Reps authenticate via Azure AD and interact in natural language — no query syntax, no system navigation. SalesBuddy retrieves relevant content from internal sources and returns a structured answer with cited source documents, so the rep can verify the information before presenting it to a customer. For complex queries, the system escalates — redirecting to named internal experts or surfacing the best available next sources, keeping humans in the loop where the situation demands it. Multimodal capabilities allow the system to process drawings and images, enabling reps to get guidance on specific application scenarios directly from technical diagrams. The interface was deployed with multilingual support, enabling reps across Castolin Eutectic's global markets to use it in their working language.
Impact
The impact of SalesBuddy is most visible in the situations it unlocks rather than the tasks it accelerates.
Reps who previously operated in narrow verticals now approach customer conversations across industrial sectors with confidence — backed by retrieval from thousands of documents covering best-practice applications, alloy selection, and proven project precedents that would have required expert consultation to access before. Cross-sector selling, which was limited by knowledge confidence, has become a realistic operational lever.
Onboarding is fundamentally different. New reps get to a productive level of technical fluency in a fraction of the previous time — not because the training program changed, but because the knowledge is now instantly accessible rather than locked inside colleagues’ heads.
Proposal preparation — previously a multi-day process for complex projects — compresses significantly. The system surfaces relevant past projects, product specifications, and application notes in a single query, giving reps a credible starting point rather than a blank page.
And perhaps most importantly: expertise that previously lived in pockets — concentrated in a small number of senior individuals — is now shared across the entire sales organization, consistently and at scale.
Looking Under the Hood
SalesBuddy runs entirely inside Castolin Eutectic’s private cloud tenant — no data leaves the enterprise boundary. The retrieval system combines semantic and keyword search with manually tuned chunk weighting to optimize precision across a technically demanding corpus. Langfuse provides end-to-end observability over every retrieval and generation step, powering a continuous evaluation loop that improves retrieval quality over time.
“We picked nexamind because they matched our way of working: honest suggestions, goals tied to business outcomes, no overpromising — and they were available side by side with us during testing and debugging, fixing fast and explaining trade-offs. The result isn't just a clever system. It's supportable, adopted, and tied to how our sales teams actually work.”
Facing similar challenges in your call center?
We'd love to understand your situation before proposing anything. Tell us what you're working on.