feat: GCP support — detection engine, LLM analyzer, and dashboard tab

[smahima27]

Summary

Adds full GCP cloud cost support to GhostBusters, based on Perforce's real SADA billing data ($359K/mo).

---

New files

gcp_detection_engine.py

Reads a SADA GCP billing CSV and runs 8 GCP-specific waste detectors:

Detector	What it catches
detect_spend_spikes	Services with >10% MoM increase (e.g. Claude Sonnet 4.5 +29%, Support +20%)
detect_sharp_drops	Services that dropped >50% — possible accidental shutdowns
detect_ai_spend	Aggregates all AI/LLM services (Claude, Vertex, Gemini) — flags spend % and growth
detect_ai_spike	Individual AI services growing faster than threshold
detect_cud_opportunity	CUD-eligible services (Compute, SQL, GKE) with low/no committed use discounts
detect_excessive_support	Support cost > 3% of total spend
detect_logging_costs	Cloud Logging > $5K/mo — review retention policies
detect_new_services	Services appearing for first time this billing period
detect_unused_savings	Large services with $0 savings programs applied

Results on Perforce SADA data:

• 37 GCP services, $359,264/mo total spend
• 20 findings, $109,417/mo additional opportunity ($1.3M/yr)

gcp_analyzer.py

Sends gcp_findings.json to Claude → produces gcp_report.json with:

• Executive summary (CTO-level)
• Ranked findings with gcp_action (GCP Console path or gcloud CLI command)
• AI/LLM spend insight
• SADA discount assessment
• Quick wins

---

Dashboard changes (dashboard_AI.py)

Added AWS / GCP tabs at the top level. GCP tab includes:

Section	Content
Metric cards	Total GCP spend, SADA savings %, additional opportunity, annual opportunity
AI Summary	Claude-generated executive summary of GCP findings
Service bar chart	Top 10 GCP services by monthly spend
Stacked savings chart	List cost vs actual spend vs SADA savings per service
MoM trend chart	All services with >5% change, color-coded red/green
AI/LLM spend panel	Breakdown of Claude, Vertex, Gemini spend with MoM % per model
Quick wins	3 GCP-specific zero-downtime actions
Findings cards	All GCP findings with severity badges and gcloud CLI fixes
SADA assessment	Claude's view on discount efficiency
GCP chatbot	Separate AI chatbot context-aware of GCP data

---

How to run

# 1. Run GCP detection
.venv/bin/python gcp_detection_engine.py

# 2. Run GCP analyzer  
export ANTHROPIC_API_KEY=$(cat SM_api_key)
.venv/bin/python gcp_analyzer.py

# 3. Launch dashboard (shows both AWS and GCP tabs)
export ANTHROPIC_API_KEY=$(cat SM_api_key)
.venv/bin/streamlit run dashboard_AI.py

Or use env vars for custom paths:

GHOSTBUSTERS_GCP_CSV="your_sada_report.csv"
GHOSTBUSTERS_GCP_FINDINGS="gcp_findings.json"
GHOSTBUSTERS_GCP_REPORT="gcp_report.json"