I build intelligent
systems as a
Machine Learning EngineerAgentic AI Systems Builder

KlearNow.AI

• Designed and deployed an AGNO-based multi-agent Document AI system using Claude 4.5 Sonnet, coordinating 6 microservices for document classification, OCR, and structured data extraction, processing 3,000+ documents/day with 99% reliability.
• Modernized document extraction workflows by migrating from LayoutLM and BERT models to an LLM-powered architecture, enabling seamless adaptation to 1000+ document templates and removing the need for template-specific retraining.
• Built a Validation Agent enforcing 14+ schema rules on LLM outputs, mitigating hallucinations, reducing downstream error rates by 30%, and cutting manual correction effort by 85%.
• Built event-driven pipelines using Apache Kafka and gRPC across 6 microservices for end-to-end processing; implemented AWS SQS-based architecture reducing queue backlog by 60% at peak load.
• Designed prompt-based extraction workflows generating structured JSON outputs from OCR text, achieving 92%+ field-level accuracy across invoice and customs formats.

KlearNow.AI

• Fine-tuned Qwen and LLaMA for structured document extraction using LoRA/QLoRA on 4-bit quantized models, reducing GPU memory by 60% and improving field-level accuracy by 18% over zero-shot baselines across 5+ document categories.
• Designed one-shot and few-shot prompting strategies for document extraction, reducing prompt token usage by 30% while maintaining extraction accuracy across 5+ document categories.
• Developed image preprocessing pipelines including orientation correction, denoising, and OCR optimization, reducing document processing latency by 35%.

Beans.ai

• Created indoor maps for 10+ facilities using ArcGIS and geospatial datasets, improving delivery routing accuracy.
• Developed a CNN-based malaria detection app with Grad-CAM visualization, achieving 94% accuracy on 27K+ blood smear images.