careerops-agent-engine

CareerOps Agent Engine is a stateful, evidence-grounded AI backend for analysing jobs, matching approved career evidence, reviewing CV changes, and generating verified DOCX/PDF CV versions. It is described as a portfolio-grade AI engineering project built around the rule that LLMs can reason and propose, but trusted state changes must be grounded, validated, persisted, and human-approved. It combines LangGraph, LangChain, LangSmith, provider-neutral LLM adapters supporting Google Gemini and Groq, FastAPI, PostgreSQL, deterministic document processing, Docker, and CI into one durable workflow.

The system includes a CV evidence pipeline that securely ingests PDF/DOCX CVs, extracts structured evidence proposals, detects overlaps, and requires explicit human approval before evidence enters the trusted registry. A job-analysis graph extracts requirements, uses a bounded tool-calling agent to discover approved evidence, calculates a deterministic fit score, generates grounded CV proposals, verifies factual claims, and pauses for human review. Verified document output applies accepted changes to a structured CV, creates an immutable version, renders an ATS-friendly DOCX, converts it to PDF with LibreOffice, and deterministically verifies both artifacts before download.

The project treats approved evidence as the source of truth, with the agent searching a user-scoped evidence registry instead of inventing experience. Agentic reasoning is bounded, and critical decisions such as fit scoring, evidence validation, review validation, versioning, artifact integrity checks, and document verification are implemented in code. Human review is a real workflow boundary, and state survives process restarts through PostgreSQL audit records and LangGraph PostgreSQL-backed checkpoints.