Make SentinelAI runnable locally via Docker Compose (golden path)

[Copilot]

The repo had no coherent local run story: broken CI filename, incomplete docker-compose.yml (Prometheus/Grafana only, missing services: key), no Dockerfiles for most services, and a README quickstart pointing at directories that didn't exist.

Core: docker compose up --build now starts the full stack

Seven services, wired with healthchecks and depends_on:

Service	Lang	Port
postgres	—	5432
ingestion-service	Go	8080
drift-engine	C++ + Python	7070
llm-guard	Python	8000
streamlit-dashboard	Python	8501
prometheus	—	9090
grafana	—	3000

New files

• .env.example — WAREHOUSE_MODE=postgres default; Snowflake vars present but blank for opt-in
• docker/initdb/01_schema.sql — bootstraps inference_logs, drift_baselines, drift_scores, incidents, incident_summaries on first Postgres start
• drift-engine/ — drift_engine.cpp now reads JSON from stdin / emits JSON to stdout; wrapped by a FastAPI server (server.py); multi-stage Dockerfile (GCC → Python slim)
• ingestion-service/main.go — proper Go module; writes to Postgres when WAREHOUSE_MODE=postgres, logs otherwise; exposes /log, /health, /metrics
• llm-guard/app.py — calls Ollama for summarization, falls back to rule-based stub if unreachable; writes summaries to incident_summaries
• streamlit-dashboard/app.py — four-tab UI (inference logs, drift scores, incidents, service health) reading directly from Postgres
• monitoring/grafana/provisioning/ — Grafana auto-provisions Prometheus datasource on startup
• .gitignore

Fixes

• Renamed .github/workflows/ci.yml) → .github/workflows/ci.yml; replaced stub content with a full pipeline (Compose build, go vet, C++ build, Python lint, pytest)
• Added permissions: contents: read to CI workflow
• README quickstart is now cp .env.example .env && docker compose up --build with curl examples

Quick example

cp .env.example .env
docker compose up --build

# ingest a log
curl -X POST http://localhost:8080/log \
  -H "Content-Type: application/json" \
  -d '{"model_id":"demo","latency_ms":120,"status":"ok"}'

# compute drift
curl -X POST http://localhost:7070/drift \
  -H "Content-Type: application/json" \
  -d '{"model_id":"demo","feature_name":"latency","expected":[0.2,0.3,0.25,0.25],"actual":[0.1,0.35,0.30,0.25]}'

Original prompt

Create a pull request to make Trojan3877/SentinelAI runnable locally via Docker Compose as the primary "golden path".

Goals

1. Running docker compose up --build from repo root should start a coherent local stack:
   • Postgres as the default local warehouse
   • Ingestion service
   • Drift engine service
   • LLM guard service
   • Streamlit dashboard
   • Prometheus + Grafana
2. Keep Snowflake integration optional (do not require it for local dev). Prefer an explicit env switch like WAREHOUSE_MODE=postgres|snowflake.
3. Fix obvious repo papercuts that prevent automation / correctness:
   • Rename workflow file .github/workflows/ci.yml) to .github/workflows/ci.yml and ensure the README badge points to the correct workflow file.
   • Ensure README quickstart matches actual repo structure and uses Docker Compose as the recommended path.

Required changes

• Add .env.example with safe defaults for local mode, including Postgres defaults and placeholders for Snowflake configuration.
• Replace/upgrade docker-compose.yml so it defines a full stack and is syntactically correct (top-level services: etc.), with sensible ports and healthchecks.
• Add Postgres initialization SQL under docker/initdb/ (or similar) to create minimal tables needed for local mode; keep schema minimal but practical for:
  • inference logs
  • drift baselines
  • drift scores
  • incidents
  • incident summaries
• Add minimal Dockerfiles for each service if missing so Compose can build:
  • ingestion-service/ (Go multi-stage)
  • llm-guard/ (Python)
  • streamlit-dashboard/ (Python)
  • drift-engine/ (C++ service). If the drift engine is currently a standalone drift_engine.cpp, move it into a drift-engine/ directory and wrap it with a tiny HTTP server (or comparable) so other services can call it.
• Add/adjust Prometheus config so it scrapes /metrics from local services.
• Ensure containers expose /health endpoints (or at least healthchecks) and Compose uses depends_on appropriately.

Documentation

• Update README.md so the first Quickstart is:
  1. cp .env.example .env
  2. docker compose up --build
  3. URLs for Streamlit/Grafana/Prometheus
  4. A simple curl example to post a sample inference log
• Update Contributing.md if needed to mention Docker Compose local dev.

CI

• Fix workflow filename and ensure CI runs at least:
  • Build docker images (or docker compose build)
  • Basic lint/test commands if already present

Notes

• The repository contains multiple languages (Python, TypeScript, C++, Dockerfile, CSS). Preserve existing structure where possible but prioritize a working Compose-based local path.
• Keep changes small and incremental; if some services are currently placeholders, provide minimal working stubs so the stack starts cleanly.

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: https://github.com/Trojan3877/SentinelAI how can this repo be improved?

Assistant: I’m the GitHub Copilot Chat Assistant. Here are the highest-impact ways to improve Trojan3877/SentinelAI based on what’s currently in the repo (default branch main, latest commit on 2026-03-17).

1) Fix repo “paper cuts” that block automation

• Rename the workflow file: you currently have .github/workflows/ci.yml) (note the trailing )). That will likely prevent GitHub Actions from running at all, and it also makes your README badge point to a file name that doesn’t exist (ci.yml).
  Action: rename to .github/workflows/ci.yml.
• Fix README quickstart paths: README says cd drift-engine, but the repo root shows drift_engine.cpp and ingestion_cpp/—not drift-engine/.
  Action: align README commands with actual directories/files.

2) Make it installable/runnable in a predictable way

Right now you have multiple components (Go ingestion, C++ drift engine, Streamlit UI, FastAPI deps, k8s/helm/terraform), but the “one command to run it” story is incomplete.

• Add a proper docker-compose.yml header: your compose file only defines services (Prometheus/Grafana) and appears to be missing the top-level services: key and version (depending on syntax).
  Action: make docker-compose up work and include core services (ingestion, dashboard, etc.) or rename it to docker-compose.monitoring.yml.
• Pin Python dependencies: requirements.txt is unpinned (e.g., fastapi, torch, transformers). This will create breakage over time.
  Action: pin versions, or adopt uv/poetry and commit a lockfile.
• Document runtime prerequisites: you mention Snowflake, SageMaker, Ollama/LangChain, Kubernetes—add an .env.example explaining required env vars and “local mode” defaults.

3) Strengthen CI quality gates (beyond “it builds”)

Your Contributing.md calls out Black + Pytest, but there’s no proof they run end-to-end.

• CI should run per-language checks:
  • Python: ruff/`black --chec...

This pull request was created from Copilot chat.