audit/22_Known_Gaps_Limitations.md

Topic 22: Known Gaps & Technical Limitations

Audit Date: 2026-02-01 Auditor: Agent Antigravity Scope: Integrity Assessment

---

1. Simulation Boundaries

While the system is "Real-World Ready", certain integrations are simulated for legal/hackathon reasons:

• Police Reporting: The system generates the JSON payload for CyberCrime.gov.in but DOES NOT actually POST it (to avoid spamming law enforcement).
• Bank API: The system validates IFSC codes mathematically but does not check if the account actually exists at the bank (requires private banking APIs).

2. Single-Channel Focus

• Current State: Optimized heavily for Text/Chat (WhatsApp/Telegram style).
• Limitation: Voice Call handling is not implemented (requires STT/TTS pipeline).
• Impact: Scams involving voice calls are not currently handled.

3. Session Persistence

• Current: In-Memory + DB Backup.
• Gap: If the container restarts suddenly (kill -9), the very last message in memory might be lost before Async commit.
• Fix: Redis implementation (Planned for V2).

4. Language Support

• Current: Strong support for English and Hinglish (Hindi-English mix).
• Limitation: Regional vernaculars (Tamil, Malayalam) rely entirely on the LLM's base capabilities without specific fine-tuning.

---

Conclusion

These gaps are Intentional Strategy Choices for the Hackathon scope, not engineering failures. The core Scam-Busting loop is unaffected.