Overview (EN)

Recruitment AI – high‑level overview¶

Recruitment AI is a web application that helps HR teams manage candidates, analyze CVs, generate personalised feedback emails, and handle incoming email inquiries using company knowledge (RAG).

Who is this for¶

• HR team – people running recruitment processes, sending feedback, answering candidate questions.
• Managers / business owners – want a clear view of the pipeline (candidates, stages, sent emails, IOD / GDPR tickets).
• Candidates – receive feedback emails and answers to their questions.

Main features¶

• Candidate management – add and edit candidates, assign them to positions, track recruitment stage.
• CV upload and parsing – upload PDF files and automatically extract key information (experience, education, skills).
• AI‑generated feedback on rejection – create constructive, personalised feedback emails based on the CV, HR notes and job description.
• Email sending – send feedback via SMTP (Zoho, Gmail, others), including consent information and privacy‑policy links.
• Inbox monitoring (optional) – read emails via IMAP, classify them with AI (IOD/GDPR, consent yes/no, normal conversation), create tickets and/or send automatic replies.
• RAG (answers from company documents) – use a Qdrant vector database to answer candidate questions based on internal documents (policies, GDPR/RODO, recruitment rules, info about the company).
• Admin panel & metrics – see candidates, sent emails, HR notes, tickets, and basic statistics/metrics.

RAG in simple words: instead of only using the “raw” AI model, the system can search company documents and feed them into the model as context (for example GDPR policies or recruitment procedures).

Flow 1: from CV to feedback email¶

Scenario: HR rejects a candidate and wants to send a clear, consistent feedback email.

sequenceDiagram
    participant HR as HR
    participant App as Application
    participant AI as Azure OpenAI
    participant Email as Mail server

    HR->>App: 1. Add candidate and upload CV (PDF)
    HR->>App: 2. Fill in HR note and choose decision \"Rejected\"
    App->>AI: 3. Parse CV (CV parser agent)
    AI-->>App: 4. Structured CV data (CVData)
    App->>AI: 5. Generate feedback (feedback agent)
    AI-->>App: 6. HTML email content
    App->>AI: 7. Validate email (validator agent)
    AI-->>App: 8. Validation result (OK or issues)
    App->>AI: 9. (optional) Correct email (correction agent)
    AI-->>App: 10. Corrected HTML
    App->>Email: 11. Send email to candidate
    Email-->>HR: 12. Delivery / log visible in admin panel

In practice, HR sees in the UI: a list of candidates, their stage, a “Process”/“Reject with feedback” action and the history of notes and sent emails.

Flow 2: from candidate email to reply or escalation¶

Scenario: a candidate replies to an email or sends a new question related to recruitment, personal data, or offer details.

sequenceDiagram
    participant Cand as Candidate
    participant Inbox as Mailbox (IMAP)
    participant Listener as EmailListener
    participant Router as EmailRouter
    participant AI as Azure OpenAI + RAG
    participant HR as HR/IOD

    Cand->>Inbox: 1. Sends an email (question / request)
    Listener->>Inbox: 2. Reads new messages
    Listener->>Router: 3. Passes email content
    Router->>AI: 4. Classify (IOD / consent / normal)
    AI-->>Router: 5. Decision: iod | consent_yes | consent_no | default

    alt IOD / GDPR
        Router->>HR: 6a. Create IOD ticket and forward to IOD/HR
    else Consent yes/no
        Router->>HR: 6b. Update consent flag for candidate
        Router-->>Cand: 7b. (optional) Confirmation email
    else Normal question
        Router->>AI: 6c. Fetch context from Qdrant (RAG)
        AI-->>Router: 7c. Suggested reply or \"forward to HR\"
        alt Confident answer
            Router-->>Cand: 8c. Send AI answer on behalf of HR
        else Low confidence
            Router->>HR: 8d. Forward email for manual answer
        end
    end

From the HR perspective:

• the system can automatically handle repetitive, simple questions (FAQ),
• all sensitive topics (GDPR, individual decisions, complaints) are still routed to a human (IOD/HR),
• AI decisions can be audited (metrics, admin panel).