Supplier Intelligence and Procurement Search

Clearlead's involvement spanned two workstreams that were distinct in nature but closely connected in outcome. The first was an oversight and advisory role: reviewing the technical architecture as it was being designed and built, and redirecting it when the initial direction would not have produced the search quality the use case required. The second was a direct implementation workstream that emerged from that oversight: defining and building the supplier data enrichment and classification system that became the data layer underpinning retrieval.

• Hybrid retrieval architecture: guided the development team from a purely semantic search approach toward a hybrid design, combining vector embeddings with structured metadata filtering and integration of transaction history from the client's operational database. Queries with specific parameters, such as brand, machine type, or part category, use metadata filters to narrow the search space first; embedding similarity handles less structured queries as a fallback. This architecture ensures procurement queries are answered with accuracy and relevance, not broad semantic approximation.

• AI agent for supplier data enrichment: built a pipeline in which an LLM agent with web search tools autonomously researched each supplier's online presence across the full 8,000-supplier base, extracting brands represented, products and services at a depth suited to downstream classification, and an initial categorisation of supplier type. The emphasis was on specificity in the extracted descriptions rather than brevity, so that structured classification could be performed downstream without re-visiting source data.

• Taxonomy for machinery parts: the classification structure was derived from how experienced staff described their manual sourcing process, then validated against their domain knowledge. The resulting multi-level hierarchy covered equipment applications, machine types, and component areas, and was used to classify the specific products each relevant supplier supplied, with multi-category assignment where a supplier's range spanned more than one area.

• AI-discovered taxonomy for supplementary product categories: for the products that fell outside the core machinery parts range, no classification framework existed. Rather than imposing a predefined structure, the enriched supplier descriptions were used to discover what product categories were actually present across the supplier base: batches of suppliers were analysed to surface category candidates, redundant and overlapping labels were consolidated, and the result was reviewed and approved by the client before being used to classify the supplementary products each supplier carried. The outcome was a structured, searchable taxonomy covering categories including safety equipment, personal protection, and consumables that had never been formally catalogued, and that previously made matching impossible even when qualified suppliers existed in the network.

• Integration with the retrieval system: the enriched and classified supplier profiles fed directly into the vector indexes and metadata filtering layer of the procurement assistant, providing the structured attributes needed for precision retrieval on technical queries.