Ichtaka AI Agent
A specialized Trust & Safety microservice providing automated PII scrubbing, digital forensics, and AI-generation detection.
01. The Problem
In an era of AI-generated misinformation and strict data privacy regulations (GDPR/NDPA), anonymous reporting systems face a crisis of trust. The challenge was to architect an automated 'Truth & Safety' layer capable of mathematically verifying the integrity of evidence while programmatically enforcing the right to privacy through PII redaction.
02. The Logic
Automated PII detection and redaction using NLP (Presidio/Spacy) and Vision (OpenCV) models.
Digital forensics through Error Level Analysis (ELA) to detect image manipulation and compression variance.
AI-generated content detection using Vision Transformer (ViT) models from HuggingFace.
Asynchronous task orchestration with Celery/Redis for high-throughput processing.
03. The Stack
04. The Solution
Implementation Result
Developed a stateless, worker-based moderation engine that analyzes incoming reports in parallel, calculating a unified integrity score used to drive automated escalation workflows.
Key Outcomes
- 01.98% accuracy in automated facial and license plate blurring
- 02.Integrated AI manipulation detection with high confidence scoring
- 03.Horizontally scalable worker architecture for high-volume report processing
Reflection
- Optimizing ML model loading for high-performance FastAPI services.
- Implementing mathematical forensics (ELA) for real-world forgery detection.
- Orchestrating multi-modal analysis (Text + Image) in a decoupled microservice architecture.