Finding suitable candidates often requires recruiters to choose between searching an internal database that may be limited or outdated, or starting a new LinkedIn search that can be slow, manual, and difficult to organise.
This use case explores how a connected AI automation can search an internal talent database first, assess candidate fit against a structured hiring request, and automatically move to live sourcing only when strong internal matches are not available.
The Problem
Filling a role often means choosing between two difficult options: searching an internal candidate database that may be thin or stale, or going straight to LinkedIn sourcing, which can be slow, manual, and unstructured.
Most recruitment tools require teams to choose one path before they know whether suitable candidates already exist internally. This can lead to unnecessary sourcing work, higher costs, and missed opportunities to use candidate information the business already has.
The Solution
The Talent Finder Agent uses a search-your-own-database-first approach.
It supports two entry modes: a chat mode for open-ended questions about the talent pool, and a search mode for structured hiring requests. Structured searches are matched against an internal Pinecone vector database using domain-aware scoring logic.
When no strong internal match is found, the workflow automatically moves to live LinkedIn sourcing. New candidate information can then be structured into resumes and added back into the database, helping the internal talent pool grow over time.
How the Agent Works
- Receive a talent request. Users can ask questions about the existing talent pool or submit a structured hiring request for a specific role.
- Search the internal database first. The workflow checks existing candidate information before starting a new sourcing process.
- Assess candidate fit. AI compares candidates against the role requirements, considering job domain, skills, seniority, experience, and location.
- Source new candidates when required. When no strong internal match is available, the workflow automatically starts a live sourcing process.
- Strengthen the talent database. Newly sourced candidate information and uploaded resumes are processed and added to the talent index for future searches.
Technical Workflow
1. Dual-mode intake
A single webhook (/talent-finder) accepts requests in two modes, distinguished by a mode field. A code node normalizes structured search requests into clean embedding-ready text.
An IF node routes mode: chat to a conversational RAG agent. Anything else proceeds down the matching pipeline.
2. Chat mode, free-form database Q&A
A separate AI agent (GPT-4.1-mini) handles open-ended questions such as “how many talents do we have?” It is instructed to always retrieve from Pinecone rather than answer from general knowledge, perform broad semantic queries for whole-database questions, and explicitly state when no data exists rather than fabricating an answer.
3. Search mode, internal database matching
The structured query first goes to a LinkedIn URL builder, used later only if needed, then to a dedicated matching AI Agent with a detailed domain-aware scoring rubric.
It groups skills and job titles into families, requires the candidate’s domain to genuinely align with the job, and scores each candidate on domain fit, must-have and nice-to-have skills, seniority, and location. The workflow returns a sorted, strictly formatted JSON list.
4. Automatic fallback to live sourcing
A code node parses the agent’s output and checks what_next. If no candidate scored ≥60, the workflow automatically triggers a live Phantombuster LinkedIn Search Export using the pre-built search URL.
If a strong match was found internally, that response is sent straight back and the LinkedIn path is skipped entirely.
5. Enriching sourced LinkedIn profiles into resumes
When fresh candidates come back from Phantombuster, each is passed to a resume-generation LLM chain that turns raw LinkedIn profile fields into a clean, ATS-optimized HTML resume.
The workflow is explicitly instructed never to invent data. This HTML is cleaned, stripped of markup, chunked into 800-character sections with 150-character overlap, and written into the same Pinecone talent index.
6. Continuous resume ingestion from Google Drive
Independently of the search flow, a Google Drive trigger watches a “CV” folder for new files every minute.
New uploads are downloaded, looped through, and fed into the same Pinecone vector store through a document loader and OpenAI embeddings pipeline, keeping the internal talent database growing passively.
Technology and Integrations
Built with: n8n, Pinecone, OpenAI GPT-4.1-mini, Phantombuster, Google Drive, Google Docs-to-PDF conversion, webhooks, and connected recruitment workflows.
Outcome
The Talent Finder Agent creates a more efficient search-then-source process for recruitment teams.
It tries fast internal matching first, using domain-aware scoring to reduce false-positive matches across unrelated fields. Only when suitable internal candidates are unavailable does the workflow move to live sourcing.
Every new candidate found through sourcing can be added back into the internal database as a structured resume, helping the talent pool become more useful over time.
Explore Custom AI Automation for Your Business
The Talent Finder Agent is one example of how AI automation can support more structured candidate search, talent sourcing, and recruitment workflows.
Divverse Labs designs and builds custom AI agents, automation workflows, internal tools, and connected systems for a wide range of business processes.
From sales, marketing, recruitment, customer support, and reporting to finance, operations, and internal team workflows, each solution is designed around the way your business works.
Request a Quote for a Custom AI Automation Solution to get started.