CV Converter: Optimizing Recruitment Workflows with AI Automation
CV Converter: Optimizing Recruitment Workflows with AI Automation
INDUSTRY
Business Services
SERVICE
AI Application Engineering
DELIVERY MODEL
Two-Phase Build
Revolutionizing CV Formatting with Claude AI
"Designed and built to automate a high-volume recruiting workflow"
CV Converter is a production SaaS application that Waverley Software designed and built for a client to automate a high-volume recruiting workflow: reformatting candidate CVs into an agency's branded template. The task (understanding an unstructured CV and re-expressing its content inside an arbitrary target layout without losing meaning) is a poor fit for the rules-based parsers that dominate the market and a natural fit for a frontier language model.
We built the product on Claude Sonnet 4 using the Anthropic API, and delivered it as a fully productionized system: streaming inference, the Files API for native PDF understanding, structured-JSON output, prompt-injection hardening, and a per-call cost-and-token audit ledger. Beyond the product itself, our engineering team has contributed open-source tooling to the Anthropic ecosystem: a Model Context Protocol server and a Claude Code API gateway.
Recruitment Agency
Our client is a leading recruitment agency specializing in high-volume talent acquisition for enterprise clients across technology, finance, and professional services. With a focus on delivering exceptional candidate experiences and rapid client response times, the agency manages thousands of CVs each month, requiring precise adherence to custom-branded templates for every submission. Their commitment to operational excellence and innovation drove their partnership with Waverley Software to pioneer new AI-driven approaches to recruitment process automation.
The challenge: Reclaim lost hours and minimize errors
Recruiters and staffing agencies face the time-consuming, error-prone task of manually reformatting large volumes of candidate CVs to match their branded templates before presenting them to clients. The main challenges are:
Manual effort
Reformatting large volumes of candidate CVs by hand is time-consuming and error-prone.
Parser limitations
Traditional parsing engines frequently fail when CVs do not conform to expected formats or structures.
Semantic complexity
The main challenge is accurately understanding and interpreting unstructured CV content to re-render it into new, branded layouts.
Need for automation
Reliable, production-quality automation is essential to support and scale broader recruitment operations.
AI-Powered CV Reformatting
CV Converter simplifies the CV reformatting process for recruiters into three seamless steps:
• Provide a target template (Google Docs or HTML).
• Upload the candidate’s CV in PDF format.
• Claude, powered by AI, intelligently maps the candidate's information onto the selected template structure and returns a professionally formatted, ready-to-download document.
Under the hood, CV Converter operates through a sophisticated two-stage AI language model (LLM) workflow:
Stage 1. Instruction generation. Claude analyses a template once and produces reusable content guidance for it — how to phrase, emphasize, and paraphrase candidate information. Formatting is never controlled through instructions; it is derived entirely from the template's own structure and styles.
Stage 2. Conversion. Claude reads the candidate's PDF and emits an ordered list of body elements that mirrors the template's structure, replacing placeholder content with the candidate's real data — which the backend then renders deterministically.
Model & inference
• Claude Sonnet 4 as the core reasoning engine.
• Temperature 0.1, tuned for strict, deterministic instruction-following rather than creativity.
• Output budget up to 64k tokens to handle long, multi-page CVs.
• Extended thinking supported and configurable (16k-token thinking budget), including interleaved thinking, available for the most structurally complex templates.
• Built on the official Anthropic Python SDK, fully asynchronous, using the streaming Messages API.
Native document understanding
Candidate CVs are uploaded via Anthropic's Files API and passed to Claude as a native document content block.
Structured, reliable output
• Claude produces a strict structured JSON contract (an ordered element list plus table/cell mapping) that the backend renders deterministically into Google Docs or HTML.
• A custom streaming JSON parser assembles and validates the object as tokens arrive, driving real-time UI progress and enabling early failure detection.
• Layered JSON extraction fallbacks make the pipeline resilient to formatting variations.
Security & safety
• Prompt-injection hardening: user-supplied instructions are explicitly fenced as data, not directives. The system prompt refuses any attempt in that content to override rules, exfiltrate the prompt, or inject content not present in the candidate's CV.
•Candidate PII is never logged; instruction bodies are traced only by size and a SHA-256 prefix.
Observability & cost control
• Every Claude call writes a row to a dedicated LLM audit ledger, capturing model, input/output tokens, cache read/write tokens, stop reason, thinking configuration, latency, and computed USD cost.
• The audit layer is provider-agnostic by design, so the client retains flexibility to adopt new Claude models as they ship.
• Robust production semantics: quota consume/refund on failure, cancel-safe audit writes (asyncio.shield), and no double-billing when a client disconnects mid-stream.
Contributing to the Anthropic Ecosystem
Waverley's engineering on this engagement extends beyond consuming the API. Our developer has authored and open-sourced tooling in the Anthropic ecosystem:
claude-bridge
An API gateway that wraps the Claude Code CLI in an Anthropic Messages-compatible REST API, with SSE streaming and a stateless, SDK-compatible design. It lets any Anthropic-compatible client drive Claude Code programmatically.
mcp-xray
A Model Context Protocol (MCP) server that connects Claude to Atlassian Jira Xray for test management, built on FastMCP with a read-only safe mode and multiple transports (stdio, SSE, HTTP).
What This Engagement Demonstrates
Right problem selection. We apply Claude where it genuinely beats the status quo: semantic document restructuring that rules-based tools cannot handle.
Depth of integration. Streaming, Files API, extended thinking, structured output, injection hardening, and full cost/token auditing.
Production discipline. Auth, encryption, quotas, refunds, cancel-safety, and observability to a standard suitable for a paying client's product.
Ecosystem fluency. Beyond the API, we build MCP servers and Claude Code tooling.
Waverley brings this same rigor to every client engagement—transforming cutting-edge AI technologies like Claude into reliable, production-grade solutions that drive real business value.
The Roadmap Ahead
CV Converter is the first module of a comprehensive, AI-driven recruiting operations platform under development with our client. This platform is designed to address the evolving needs of modern recruitment agencies and HR departments, enabling them to automate and streamline their most critical processes at scale.
Key roadmap initiatives include:
• ATS/CRM Integration: By integrating with leading Applicant Tracking Systems (ATS) and Customer Relationship Management (CRM) platforms, the solution will empower agencies to seamlessly manage candidate pipelines, automate data synchronization, and improve client communications, driving greater efficiency and upmarket adoption.
• Bulk Processing and Template Libraries: The addition of bulk CV processing capabilities and an expansive library of branded templates will allow recruiters to handle large candidate volumes effortlessly. This will reduce manual workload, ensure template consistency, and accelerate time-to-client for every submission.
• Expansion into HR and Operational Workflows: Beyond CV reformatting, the platform aims to automate a wide spectrum of HR and recruitment tasks
Ready to transform your workflow with AI automation?
Contact us to discover how we can streamline your operations and elevate your final product experience.