Multi-agent hiring platform · QualiZeal internship
Built during an AI engineering internship at QualiZeal: an autonomous LangGraph platform that replaces a traditional applicant tracking system. It parses job descriptions and résumés, scores every candidate with cited evidence, flags gaps, writes tailored interview questions, ranks the pool and drafts outreach.
Highlights
- Six specialized agents with schema-validated outputs, coordinated through parallel map-reduce with per-candidate failure isolation.
- A human-in-the-loop gate (LangGraph interrupts): recruiters approve or edit weights and re-rank instantly without re-running the LLM.
- FastAPI with live streaming, a Next.js dashboard, and one Postgres + pgvector database for RAG, memory, checkpointing and audit logs.
- A provider-neutral LLM layer — OpenAI, Anthropic, AWS Bedrock, Google Vertex, Ollama — so models can be switched without code changes.
- Safety by design: prompt-injection and hidden-text detection, PII redaction and encryption, protected attributes kept out of scoring.
- 82 automated tests in CI: shortlists matched human recruiters on a labeled golden set, and swapping names and genders changed no scores.
Built with
LangGraphMulti-agentRAGHuman-in-the-loopPythonFastAPINext.jsPostgreSQLpgvectorGitHub Actions
Next project
MentoLabAdaptive AI coding tutor