← Role tracks

R5 Stage D · 160 h (48 T / 112 P)

Data & RAG Engineer

Builds and operates the data and retrieval layer that AI systems depend on.

Tech stack

PythonSQLAirflowdbtUnstructuredpgvectorQdrantOpenSearchsentence-transformersGreat Expectations

Modules

M1 · Ingestion at scale

40 h
TOPICS
  • Source connectors
  • Batch versus streaming
  • Change data capture
  • Incremental and delta sync
  • Complex document parsing: tables, forms, scans
  • OCR quality
  • Deduplication
  • Idempotent loads
  • Backfill strategy

Lab: Production ingestion pipeline over a heterogeneous, frequently changing corpus

Course material for this module is in production.

M2 · Transformation and modelling

40 h
TOPICS
  • Dimensional modelling
  • Transformation frameworks
  • Testing and data contracts
  • Data quality dimensions
  • Lineage and cataloguing
  • Schema evolution
  • PII handling in pipelines

Lab: Tested transformation layer with enforced data contracts and lineage

Course material for this module is in production.

M3 · Retrieval engineering

44 h
TOPICS
  • Embedding model selection and benchmarking
  • Chunking experimentation
  • Index type and parameter tuning (HNSW/IVF)
  • Metadata filtering
  • Hybrid search fusion
  • Reranking
  • Freshness and TTL
  • Multi-tenancy and access control in retrieval

Lab: Tune an index to meet stated recall and p95 latency targets

Course material for this module is in production.

M4 · Operating the retrieval layer

36 h
TOPICS
  • Retrieval evaluation harness
  • Monitoring recall drift
  • Corpus growth and re-embedding strategy
  • Cost of storage and embedding
  • Capacity planning
  • Disaster recovery
  • Corpus poisoning defence

Lab: Retrieval evaluation and monitoring suite with scheduled refresh

Course material for this module is in production.

Track project

Production retrieval platform: ingestion of a multi-format changing corpus, tested transformation layer, tuned hybrid index with reranking, access-controlled multi-tenant retrieval, evaluation harness reporting recall@k, MRR and faithfulness, plus scheduled refresh and monitoring.

Job-ready exit standard

Builds and measures a retrieval layer; defends chunking, embedding and index choices with numbers.