AI Agents
CHAPAL
An AI safety and auditing platform that improves trust and reliability in LLM-based systems through dual-layer threat protection and human-in-the-loop governance.
Tech Stack
Next.js
TypeScript
TailwindCSS
Prisma
Gemini
The Problem
- LLM deployments had no robust defences against prompt injection, PII leakage, and real-time policy violations.
- Hallucinations, unsafe advice, and toxic outputs from LLMs created serious production reliability risks.
- No unified auditing layer gave administrators visibility into flagged AI responses before reaching end users.
- Teams lacked a structured feedback loop for continuously improving AI safety policies from real interaction data.
- Testing adversarial scenarios required manual effort with no repeatable developer simulation tooling.
Gallery
Our Solution
- Implemented a deterministic guard layer for high-speed mitigation of prompt injection and PII leakage threats.
- Built a semantic auditing layer using Llama 3.1 via Groq and Google Gemini to detect hallucinations and unsafe content.
- Designed a Human-in-the-Loop workflow enabling administrators to review, block, or rewrite flagged AI responses.
- Created transparent analytics dashboards covering safety scoring and sentiment tracking across all interactions.
- Developed developer simulation tools for testing adversarial prompts and validating safety policy improvements.
Impact
Dual-layerLLM protection
Delivered a production-grade LLM governance system enabling continuous safety improvement through human oversight and automated dual-layer protection.
Ready to build something similar?
Let's discuss your project and see how we can help.