MaplePath
Production AI Express Entry assistant: LangGraph multi-step orchestration, hybrid search (BM25 + pgvector + Cohere), and deterministic CRS scoring
An intelligent, auditable immigration assistant for Canada's Express Entry skilled immigration system. Converts applicant natural language into validated profile schemas, retrieves and classifies NOC 2021 occupation codes with hybrid search, and deterministically computes Comprehensive Ranking System (CRS) points and FSW/CEC/FST eligibility.
- Architected an event-driven LangGraph orchestrator with typed state routing applicant intake through extraction, NOC retrieval, classification, and eligibility evaluation.
- Built a hybrid search pipeline combining BM25 keyword search, pgvector semantic embeddings, Reciprocal Rank Fusion (RRF), and Cohere reranking.
- Benchmarked NOC retrieval pipelines against ground-truth labeled datasets in noc/evaluate.py, evaluating hit rates and NDCG@10 scores.
- Engineered a 100% deterministic, auditable rule engine with Pydantic for CRS points, language conversions (IELTS, CELPIP, PTE, TEF, TCF to CLB), and FSW 67-point grid checks.
- Shipped a responsive Next.js frontend with live interactive CRS simulation and NOC code exploration.