Description
We're looking for an experienced agentic AI developer to build a proof-of-concept system for legal and performance reporting, working against a locally hosted structured dataset (SQLite) as the primary data source. The system should support two modes of interaction: 1. On-demand requests — the user provides an identifier (accepted either as natural language or structured input), and the system returns a complete package: a generated report, a current status assessment with supporting rationale, and scheduling context for the next review cycle. Output must be delivered as a downloadable document (PDF or Word), not just a chat response. 2. Scheduled updates — the system automatically compiles a portfolio-wide summary on a monthly basis without requiring user input, re-running its full data retrieval and reasoning pipeline live on each execution (no cached or replayed results). Core requirements: - Multi-agent architecture (LangGraph preferred), with an orchestrator agent routing to specialized sub-agents (report drafting, scheduling recommendations, status classification and recommended-action planning) and a final verification step before any output is returned - Retrieval-augmented generation (RAG) over free-text fields in the dataset plus a short supplementary reference document, to support citation-backed outputs - SQL-based data access against a local database, with automatic detection of input format - Every generated statement must cite the specific underlying record(s)