How anhtuans-project/track2_day28_2a202601582_haanhtuan works
Day 28 is an educational integration lab for an AI platform. Its main workflow accepts a request through Envoy and FastAPI, publishes ingestion events to Kafka, processes them with Airflow and Spark into Delta, and serves answers using Feast features, Qdrant retrieval, MLflow release metadata, and vLLM inference. The repository also provides readiness, telemetry, and metrics capabilities. Some infrastructure and orchestration components are represented by their documented roles; their source was not sampled, so those links cite the README rather than implying inspected implementation details.
Main entry points: pipeline.py Serving orchestration · api.py FastAPI · readiness.py Readiness probes · event_bus.py Kafka event bus
Components 20
Ingestion & orchestration
- Envoy gateway
gateway - Platform contracts
src/lab28_platform/contracts.py - Kafka event bus
src/lab28_platform/event_bus.py - Airflow pipeline
airflow/dags/lab28_ingestion_pipeline.py
Data & retrieval
- Spark Delta merge
spark/delta_merge.py - Delta lakehouse
src/lab28_platform/delta_store.py - Feature export
spark/feature_export.py - Feast definitions
feature-repo/definitions.py - Feast feature store
src/lab28_platform/feature_store.py - Qdrant retrieval
src/lab28_platform/vector_store.py
Serving & releases
- FastAPI
src/lab28_platform/api.py - Serving orchestration
src/lab28_platform/pipeline.py - MLflow registry
src/lab28_platform/model_registry.py - vLLM client
src/lab28_platform/llm_client.py - Guardrails
src/lab28_platform/guardrails.py
Readiness & observability
- Readiness probes
src/lab28_platform/readiness.py - OpenTelemetry
src/lab28_platform/telemetry.py - Prometheus metrics
src/lab28_platform/metrics.py - Monitoring stack
monitoring
Other components
- User
Connections 23
- User → Envoy gateway: sends request
README.md - Envoy gateway → FastAPI: routes requests
README.md - FastAPI → Kafka event bus: publishes events
README.md - Kafka event bus → Airflow pipeline: feeds pipeline
README.md - Airflow pipeline → Spark Delta merge: runs merge
airflow/dags/lab28_ingestion_pipeline.py - Spark Delta merge → Delta lakehouse: merges data
spark/delta_merge.py - Delta lakehouse → Feature export: provides data
README.md - Feature export → Feast definitions: exports features
README.md - Feast definitions → Feast feature store: defines features
- Serving orchestration → MLflow registry: resolves release
src/lab28_platform/pipeline.py - Serving orchestration → Feast feature store: reads features
src/lab28_platform/pipeline.py - Serving orchestration → Qdrant retrieval: retrieves sources
src/lab28_platform/pipeline.py - Serving orchestration → vLLM client: requests completion
src/lab28_platform/pipeline.py - Serving orchestration → Guardrails: screens request
src/lab28_platform/pipeline.py - FastAPI → Serving orchestration: serves through
src/lab28_platform/api.py - FastAPI → Platform contracts: validates payloads
src/lab28_platform/api.py - vLLM client → User: returns answer
- Readiness probes → Kafka event bus: probes health
src/lab28_platform/readiness.py - Readiness probes → Delta lakehouse: probes health
src/lab28_platform/readiness.py - Readiness probes → Feast feature store: probes health
src/lab28_platform/readiness.py - Serving orchestration → Prometheus metrics: records timings
src/lab28_platform/pipeline.py - OpenTelemetry → Monitoring stack: exports telemetry
README.md - Prometheus metrics → Monitoring stack: feeds metrics
README.md