```json
{
  "entry_points": [
    "app/main.py: FastAPI application bootstrap and route registration",
    "app/proxy.py: Primary API endpoint handler for /v1/chat/completions",
    "app/reports.py: Dashboard and reporting route handlers"
  ],
  "data_flow": [
    "Inbound LLM requests -> proxy.py intercepts /v1/chat/completions -> metadata extracted -> store.py writes to SQLite -> request forwarded to upstream llama.cpp -> response streamed back to client",
    "Dashboard requests -> reports.py queries SQLite via store.py -> aggregates run/request stats -> redaction.py sanitizes payloads/secrets -> formatted data returned to frontend"
  ],
  "risks": [
    "app/proxy.py: SSRF vulnerability if upstream target is dynamic/user-controlled; missing input validation/rate limiting; potential prompt injection passthrough",
    "app/store.py: SQL injection if dynamic queries are used; SQLite concurrency limits under high load; plaintext storage of sensitive prompts/responses",
    "app/redaction.py: Incomplete or regex-based secret filtering may leak API keys/tokens; bypass risks with encoded, obfuscated, or split payloads",
    "app/reports.py: Unoptimized aggregation queries can cause SQLite lock contention or DoS; potential XSS if aggregated strings are rendered without proper escaping"
  ]
}
```