Recruiting teams are sitting on more data than ever — candidates in the ATS, clients in the CRM, placements in finance, bench availability in spreadsheets. The problem is not a shortage of data. The problem is that none of it talks to each other, and getting a unified answer means logging into multiple systems, exporting files, and assembling the picture manually. Agentic AI can close that gap, but only when the siloed data underneath it is actually connected.
This guide walks through the Talent Intelligence Platform: an agentic, natural language recruiting intelligence layer built using Claude Code, with connectivity powered by CData Connect AI.
CData Connect AI acts as the managed connectivity layer — handling authentication, schema translation, and governance across every source, so development time goes toward building intelligence, not integrations. The full source code is on GitHub.
Home view of the Talent Intelligence Platform — skills in demand, placements trend, and bench visibility across all connected sources.
Why recruiting AI projects stall before they deliver value
Most recruiting organizations operate across five or more disconnected systems — ATS, CRM, HRIS, ERP, and spreadsheets. Getting a unified answer means logging into each one, exporting, and reconciling manually.
When a recruiter needs to answer “Who is available in Chicago this week with Java experience under $140K?” the data exists — it is just spread across three systems with no way to query them together. The typical result is a shortlist that took 20 to 40 minutes to build.
71% of AI teams spend more than a quarter of their implementation time on data integration alone, per CData's AI Accuracy whitepaper. Most recruiting AI projects stall here before delivering any value. Native ATS search compounds it further — keyword-only, no skill synonyms, no cross-system visibility.
The bottleneck is not the model. It is the data layer underneath it.
How CData Connect AI unifies recruiting data across every system
Building cross-system AI on top of recruiting data without a managed connectivity layer means writing a custom connector for every source, maintaining separate authentication flows per system, building a normalization layer to reconcile mismatched schemas, and layering governance on top — all before any intelligence is written.
CData Connect AI compresses that into a single governed MCP endpoint. Every connected source — Bullhorn, Salesforce, Workday, Snowflake, SharePoint, Google Drive — is accessible through standard SQL queries against one interface. Authentication, schema translation, and access governance are handled at the Connect AI layer, not inside the application.
46% of organizations require real-time access to six or more data sources for a single AI use case. For recruiting, that is the standard case, not the exception.
Concern | Without Connect AI | With Connect AI |
Data source access | Custom connector per system, separate auth flows | Standard SQL via single MCP endpoint |
Schema normalization | Manual field mapping across every source | Handled at the Connect AI layer |
Access governance | Build and maintain a custom authorization layer | Identity passthrough, RBAC, configured once |
Maintenance | Update integrations when source APIs change | CData maintains the connectors |
Time to first query | Weeks to months | Minutes |
Three capabilities made Connect AI the right choice for this platform:
Connectivity: hundreds of pre-built connectors to SaaS platforms, databases, APIs, and file systems. Schema discovery runs at runtime, so field naming inconsistencies across sources are resolved automatically before any query reaches the application.
Context: Semantic intelligence that understands the meaning behind queries. Pushdown execution runs joins and filters at the source, reducing token overhead and keeping query costs low. The entire platform runs at under $0.01 per session.
Control: Identity passthrough ensures the AI can only access data the authenticated user is permitted to see. Every query is logged with user, timestamp, source, and result — a complete audit trail across all connected systems.
What the platform does: core recruiting use cases
The platform is a natural language interface over every connected recruiting data source. Type a question and the agent pipeline queries all relevant systems simultaneously, ranks results by relevance, and explains the match.
A query like “Find Java developers with at least 5 years of experience” returns ranked candidates across all connected sources, with an AI Insights panel explaining the ranking and flagging skill gaps. The source filter tabs at the top show which systems contributed to the results in real time.
Drilling into a candidate shows full contact details, skills, compensation, availability, and a Candidate Journey pipeline tracking each stage from sourced through screening, submission, interview, offer, and placement — assembled from whichever system holds each data point.
The platform supports five core use cases out of the box:
Use case | Who uses it | Example query |
AI candidate matching | Recruiter | “Find senior Java engineers in Chicago available immediately under $140K” |
Bench optimization | Recruiting manager | “Who is available this week by skill and location?” |
Client fit intelligence | Account manager | “Find candidates who previously worked at companies similar to JP Morgan” |
Skills gap and demand analysis | Executive leadership | “What skills are most requested this quarter vs what we have on the bench?” |
Recruiter performance analytics | Director of recruiting | “Which recruiters have the highest placement rate this quarter?” |
The same data layer that powers recruiter search powers an analytics dashboard for leadership.
Analytics dashboard view — skills in demand, placements trend, and bench visibility, drawn live from every connected source.
Inside the build: how Connect AI powers the app architecture
The development workflow changes fundamentally when Connect AI handles the connectivity layer. The Talent Intelligence Platform was built using Claude Code as the AI coding assistant, with Connect AI as the data layer.
The two complement each other directly: Connect AI removes the hardest part of cross-system development — the connector plumbing, schema wrangling, and auth flows — so Claude Code can focus entirely on application logic. What would normally be weeks of integration work collapses into prompts focused on behavior.
For example, schema discovery across all connected sources required no manual mapping. A single prompt to Claude Code — “discover available tables and columns from the Connect AI endpoint at startup” — produced a tiered cache that resolves field names across catalogs at runtime.
The agent pipeline, LangGraph nodes, and rate-limiting system were all built iteratively: describe the behavior, review the output, refine. Each phase built on the last without rewriting the plumbing underneath.
Architecture flow — REST for deterministic queries, MCP for natural language, with LangGraph orchestrating intent classification, schema resolution, query execution, and analysis.
The architecture reflects two key design decisions made during development:
REST for deterministic queries, MCP for natural language. The sidebar candidate list, profile pages, and schema discovery all use Connect AI’s REST API — fast, predictable, rate-limit-friendly. The MCP endpoint is reserved exclusively for LLM-generated natural language queries. This separation keeps rate limit budgets independent and latency low for the high-frequency operations that run on every page load.
Client-side credential encryption. Credentials are encrypted in the browser using AES-256-GCM before they leave the device. They travel as HTTP headers, are never written to the server filesystem, and are never visible to the LLM. LLM credentials have no path to the data connection, and data connection credentials have no path to the LLM.
The agent pipeline runs on LangGraph: intent classification → schema resolution → query construction → execution via Connect AI → analysis and ranking. LLM providers are pluggable — Groq (free tier), Google Gemini, DeepSeek, Mistral, and OpenAI all work out of the box.
Token usage and query cost are visible in the logs dashboard. A full session of natural language candidate searches — 224,000 tokens processed — costs $0.00 on Groq’s free tier.
How to build this platform with Claude Code and Connect AI
Connect AI removes the hardest part of cross-system development — the connector plumbing, schema wrangling, and auth flows — so Claude Code can focus entirely on application logic. What would normally take weeks of integration work collapses into prompts focused on behavior.
Run the platform from the existing repo
The full source code is available at github.com/CDataSoftware/talent-intelligence-platform. The repository includes setup documentation, environment variable templates, and a connection guide for linking your own CData Connect AI sources.
Clone the repository and run npm install
Copy .env.local.example to .env.local and add your Connect AI credentials and LLM API key
Log in to cloud.cdata.com, add your data sources, and note the connection names
Run the quick setup interactive script: start.bat (Windows) or ./start.sh (Mac/Linux)
Alternatively, run npm run dev to start the server manually, then open http://localhost:3000
Build your own version from scratch with Claude Code
The repository includes a CLAUDE.md file that Claude Code reads automatically on startup. It covers the full architecture, constraints, modification playbooks, and a step-by-step recreation guide for adapting the platform to a different data source or domain. To build a version against your own data:
Answer the intake questions in CLAUDE.md (data source, LLM provider, domain, skill level)
Follow the build order: auth store → crypto → API client → REST client → schema cache → field resolver → agent nodes → UI
Connect your own source in CData Connect AI and point the endpoint at it — schema is auto-discovered, no hardcoded table names to change
Use Claude Code iteratively: describe each layer’s behavior, review, refine, and move to the next
The approach works for any domain where decisions span more than one data system.
Frequently asked questions
Can I connect my existing ATS or CRM to this platform?
Yes — CData Connect AI supports 350+ enterprise data sources including Bullhorn, Salesforce, Workday, and more. Add a connection in Connect AI and the platform auto-discovers the schema at runtime, with no manual field mapping required.
Do I need to know SQL or write custom connectors to build this?
Not necessarily. CData Connect AI handles the connectivity layer and Claude Code generates application logic from prompts. SQL knowledge helps for custom queries but is not required to get started.
How much does it cost to run natural language queries at scale?
Using Groq’s free tier, a full session of 224,000 tokens costs $0.00. The platform is designed to run at under $0.01 per session across paid LLM providers as well.
Can I adapt this platform for a domain other than recruiting?
Yes. The repository includes a CLAUDE.md build guide for adapting the platform to any domain where decisions span multiple data systems — HR, finance, sales operations, and more. Schema is auto-discovered, so no hardcoded table names need to change.
Build your next recruiting tool with CData Connect AI
CData Connect AI gives your AI agents and coding assistants governed, real-time access to 350+ enterprise data sources — so you spend time building intelligence, not integrations.
Start a free trial or book a live demo to see Connect AI running against your own sources.
Your enterprise data, finally AI-ready.
Connect AI gives your AI assistants and agents live, governed access to 350+ enterprise systems -- so they can reason over your actual business data, not just what they were trained on.
