FAQ

How does Fact AI Lab integrate with our existing LLM setup?

Fact AI Lab sits as a thin verification layer between your application and your LLM provider. We offer a Python SDK, TypeScript SDK, and a REST API that works with any OpenAI-compatible endpoint. You don't change your prompts or your model - you pipe outputs through us before they reach users. Average integration time for a single workflow is under 2 hours.

What hallucination detection accuracy do you achieve?

On SEC filing summarization and Q&A workflows, we measure under 0.3% hallucination rate after verification - compared to 4-8% unverified on the same tasks. Detection works by cross-referencing outputs against your provided source documents using a combination of embedding similarity and fact-checking probes. Results vary by task type; we measure your baseline during the pilot.

What does the audit log actually contain?

Each log entry includes: the original prompt (sanitized or full, your choice), the retrieved context chunks with source citations, the model version and temperature, the raw LLM output, the verification score, any flagged claims, and a cryptographic hash linking to the previous entry. Logs are tamper-evident and can be exported as PDF or JSON.

Is our data used to train your models?

No. Your data is never used for training. Fact AI Lab operates as verification infrastructure - we process your outputs in-flight and log the results to your designated storage (your cloud account, not ours). We don't retain your prompts or outputs beyond the 24-hour processing window unless you configure longer retention.

Which regulations does the audit trail support?

Our report templates are structured for SEC AI governance guidance (2023 and 2024 staff bulletins), FINRA's AI examination framework, FCA's Consumer Duty and AI principles, and OSFI's B-10 guidelines. HIPAA audit requirements are also supported. We don't certify compliance - that's your counsel's job - but we give them the documentation package they need.

What does pricing look like?

Pricing is usage-based plus a platform fee. We start with a 30-day structured pilot at a fixed low cost so you can measure value before committing. After the pilot, pricing scales with verified request volume. We don't publish a rate card because the right structure depends on your workflow volume and data residency requirements. Book a call and we'll give you a number within 24 hours.