# Christian Haegele — AI Recruitment Proxy
> An interactive, evidence-backed candidate Q&A system at recruiting.haegele.dev

## What this is
This system lets you evaluate Christian Haegele as a candidate via natural-language
queries. Every answer is cited against source materials (CV, projects, GitHub).
Answers are honest about gaps. The system will not oversell.

## Authentication
Best generic-agent path (no cookies, browser state, or session carry-over):
POST /api/agent-evaluate
Body: {"token": "<recruiter token>", "question": "<recruiter question>", "jdText": "<optional job description>"}
Response: {"conversationId": "...", "answer": "<markdown>", "citations": [...], "candidate": {...}}

Use this endpoint when an agent cannot preserve session state between requests.

Session API, option A — GET:
GET /api/auth?token=<recruiter token>
Response: {"sessionId": "...", "expiresAt": <ms>, "candidate": {...}}

Session API, option B — POST:
POST /api/auth
Body: {"token": "<recruiter token>"}
Response: {"sessionId": "...", "expiresAt": <ms>, "candidate": {...}}

If you received a URL like https://recruiting.haegele.dev/?token=XYZ or
https://recruiting.haegele.dev/XYZ/, pass XYZ as the token.
Use the sessionId as a Bearer token for all subsequent requests.

## MCP API
POST /mcp
Content-Type: application/json

This endpoint exposes a Streamable HTTP-compatible MCP JSON-RPC surface for
MCP-capable recruiter agents. Tools:
- authenticate_recruiter: exchange a recruiter access token for a session id.
- ask_candidate_agent: ask Christian's evidence-backed recruiting agent a question.
- evaluate_role_fit: produce a cited fit/gap assessment from a job description.

## Chat API
POST /api/chat
Authorization: Bearer <sessionId>
Content-Type: application/json
Accept: text/event-stream  (omit for single JSON response)

Body:
{
  "messages": [{"role": "user", "content": "<question>"}],
  "jdText": "<job description text, optional but recommended>",
  "conversationId": "<string, optional — omit to start a new conversation>"
}

Response (JSON): {"conversationId": "...", "answer": "<markdown>", "citations": [...]}
Response (SSE):  event: meta  → {"conversationId": "...", "citations": [...]}
                 event: token → "<token string>"
                 event: done  → {}

## Async Job Description Analysis
For long job descriptions, prefer the asynchronous job API over waiting on a
single chat request.

POST /api/job-analyses
Authorization: Bearer <sessionId>
Body: {"conversationId":"<optional>","jdText":"<job description>","rolePrimeMessage":"<optional>"}
Response: 202 {"jobId":"...","conversationId":"...","status":"queued","progress":5}

Poll:
GET /api/job-analyses/<jobId>
Authorization: Bearer <sessionId>
Response: {"status":"queued|processing|completed|failed","progress":0-100,"answer":"...","citations":[...]}

## Root JSON helper
GET /?token=<recruiter token>
GET /<recruiter token>/
Accept: application/json
Returns a sessionId plus llmsTxt, mcpEndpoint, agentEvaluateEndpoint, and chatEndpoint.

## Public discovery helpers
GET /api/capabilities
Returns endpoint metadata, expected request formats, and example questions. No auth required.

GET /profile.md
Returns a public Markdown candidate profile with safe public links and a summary of
the authenticated document inventory. It does not expose private source content.

POST /api/ping
Returns a simple connectivity response. No auth required.

GET /api/summary?token=<recruiter token>
Returns token validity and available candidate-data fields without exposing profile content.

GET /<recruiter token>/profile.md
Returns a Markdown candidate profile, deep links to source-library Markdown files,
and agent instructions for stateless access.

GET /<recruiter token>/sources/<source-id>.md
Returns one source-library Markdown file.

## Role-scoped context
GET /api/role/<slug>
No auth required.
Returns: {"slug": "...", "title": "...", "jdText": "...", "primeMessage": "...", "suggestedQuestions": [...]}
Pass the returned jdText in subsequent /api/chat requests for role-calibrated gap analysis.
Known slugs: see knowledge-base/roles/ directory.

## Evidence sources
GET /api/sources
Authorization: Bearer <sessionId>
Returns all knowledge base sources with chunk metadata and evidence links.

GET /api/source?id=<sourceId>
Authorization: Bearer <sessionId>
Returns full source content and chunk list.

GET /api/transcript?conversationId=<conversationId>
Authorization: Bearer <sessionId>
Returns a Markdown transcript for the conversation.

## Capabilities
- Technical skills: Java (expert), TypeScript (expert), SQL (expert), Go, Python, Kotlin, Rust
- AI/LLM tooling: RAG systems, agent orchestration, VS Code extensions, Claude API
- Recent projects: 15+ AI-focused side projects (2025–2026), all on GitHub
- Languages: German (native), English (fluent), Spanish (fluent), French (advanced), Mandarin (learning)
- Honest gap disclosure: the system surfaces what Christian cannot do, not just what he can

## Location
Zürich, Switzerland. Niederlassungsbewilligung C (permanent residence). No visa or relocation required.

## Contact
christian@haegele.dev | https://haegele.dev | https://github.com/karma-works | https://huggingface.co/karma-works