Responsible AI & Operational Compliance Policy
Last Updated: November 18, 2026 · Architectural Standard for Production Conversational Systems
1. Strict Knowledge Grounding & Anti-Hallucination
AUTOLEZ is engineered specifically for mission-critical commercial workflows including real estate inquiries, legal intakes, medical appointments, and retail checkouts. To prevent hallucinations and false claims, our platform adheres to strict Retrieval-Augmented Generation (RAG) principles:
- Document-Restricted Answering: AI assistants answer questions exclusively from verified documents (brochures, rate sheets, service schedules, and FAQs) uploaded by the firm's managing team.
- Explicit Uncertainty Handling: If a customer query cannot be answered with high confidence from verified business records, the AI is programmed to transparently acknowledge that the information is unavailable and offer to connect the customer with a human staff member rather than speculating.
- Vector Embedding Search: Knowledge chunks are indexed via high-dimensional vector embeddings, ensuring semantic precision and eliminating out-of-context retrieval.
2. Human-in-the-Loop & Instant Takeover
Autonomous AI is an operational multiplier, not a replacement for human judgment. AUTOLEZ incorporates multiple layers of human supervision:
- 1-Click Human Takeover: Inside the Unified Inbox, any authorized team member can toggle "Human Takeover" on any thread at any time. When activated, the AI immediately halts automated message generation on that conversation thread.
- Internal Team Notes: Staff can collaborate internally with threaded private notes on any lead record without exposing draft messages to the customer.
- Transparent Conversation Logs: All customer interactions across WhatsApp, Instagram, Facebook, and Webchat are recorded with timestamps, channel identifiers, and complete message histories.
3. Multi-Tenant Data Isolation & Privacy
Customer data confidentiality is foundational to the AUTOLEZ architecture:
- PostgreSQL Row Level Security (RLS): Every workspace's leads, documents, and transcripts are partitioned at the database layer. No workspace can read or modify data belonging to another tenant.
- No Model Retraining on Customer Data: Customer conversations, proprietary documents, and personal details are strictly isolated. We do not use customer data to train or fine-tune public base models.
- DPDP Act 2023 Alignment: Data handling practices comply with the India Digital Personal Data Protection Act 2023 and internationally recognized privacy norms, supporting data access requests and account deletion.
4. Circuit Breakers, Spend Controls & Rate Limits
To protect against runaway usage, denial-of-wallet attacks, and system failures, AUTOLEZ enforces multi-tiered platform circuit breakers:
- Global Emergency Kill Switch: Platform overseers can activate an emergency circuit breaker that instantly terminates public LLM and Realtime Voice sessions platform-wide or per-tenant.
- Atomic Quota Reservations: AI sessions and voice minutes are atomically reserved before requests reach LLM or SIP gateways, preventing concurrency spikes from exceeding monthly quotas.
- Distributed Rate Limiting: Public intake endpoints (e.g. website chat, phone webhooks) enforce fail-closed distributed IP and tenant throttling to block abusive automated traffic.
5. Inquiries & Compliance Contact
For questions regarding our AI governance, responsible operational practices, or security audits, contact our compliance team: