HRTech Startup Transforms Vendor-Locked Skills Data into an Owned AI Taxonomy Engine with Claude on AWS Bedrock
At a glance
An HRTech startup building an AI-driven career platform for cloud, AI, and cybersecurity professionals depended entirely on a third-party vendor for skill extraction. That vendor’s taxonomy updated too slowly to keep pace with emerging AI-era skills, undermining the platform’s core promise to its own users.
NMD designed and built a proprietary, Claude-powered skills-intelligence platform on AWS, replacing the vendor black box with an owned taxonomy, extraction, and matching engine. The direct outcome: the startup now controls its own skills data asset instead of renting someone else’s.
The startup converted a recurring vendor dependency into an owned, extensible AI data platform, and used that same foundation to open a second product line evaluating AI agents as “digital professionals.”
Industry
Use Case
Intelligent Document Processing, Knowledge Management, Personalization & Recommendation
Solution implemented
- Claude on Amazon Bedrock powers skill and profession extraction from resumes and job postings.
- Retrieval-augmented generation constrains extraction to a client-defined, constrained taxonomy.
- Amazon RDS for PostgreSQL with vector search stores and serves the skills and profession ontology.
- AWS Lambda and API Gateway expose extraction, normalization, and autocomplete as API services.
- AWS Glue supports batch ingestion and embedding of large historical candidate datasets.
- Weighted must/should/could scoring and LLM-based letter-grade rationale drive candidate ranking.
The value equation
- Replaces a rented, black-box vendor taxonomy with an owned, extensible AI data asset.
- Extraction now covers emerging AI-era skills the incumbent vendor could not track.
- Positions the platform to launch a new agent-evaluation product line.
- Removes recurring per-token, per-subscription vendor cost exposure over time.
Company Snapshot
An early-stage HRTech startup providing AI-driven career and job-matching tools for cloud, AI, and cybersecurity professionals.
Location
United States
Customer Situation
The startup’s product depends on matching technology professionals to opportunities based on their real, current skills. Its incumbent vendor’s taxonomy was closed, updated on a slow cadence, and had already missed skills core to the startup’s own positioning, including tools and techniques specific to agentic AI work.
This was more than an accuracy gap. It meant the platform’s most differentiated users, the earliest adopters of new AI tools, were the ones least likely to be correctly represented in the system meant to showcase them.
The broader problem is industry-wide: legacy HR-tech taxonomies were not built for a market where new AI skill categories can emerge and become market-relevant within months.
NMD Solution
NMD reviewed the incumbent vendor integration and found that the core constraint was not the API layer but the underlying taxonomy: closed, slow to update, and not owned by the client. The solution required a constrained, RAG-based extraction pipeline built on Claude via Amazon Bedrock, backed by a client-owned skills and profession ontology stored in a vector-enabled PostgreSQL database on AWS, with API Gateway and Lambda exposing extraction, normalization, and search.
What We Delivered
NMD designed and built a proprietary skills-intelligence platform that replaces the client’s third-party vendor dependency: constrained skill and profession extraction, normalization against an owned taxonomy, autocomplete search, and a skill-to-profession ontology, deployed on AWS with Claude as the extraction and grading engine. The build extends into a companion capability evaluating AI agents against the same skill and capability framework used for people, opening a second product surface for the platform.
Ready to Transform Your Software Business?
See how New Math Data can transform your software business with AWS-powered innovation.