Yash

Built reliable AI systems and contributed to FailproofAI (secure AI guardrails) and Claudeye (tmux session monitor overlay for Claude Code). Developed hook-based blocker frameworks that prevent illegal or unsafe agent actions in real-time.

• →Built Claudeye (tmux session monitor overlay)
• →Contributed to FailproofAI guardrail runtime
• →Developed hook-based action blocker frameworks
• →Created adaptive guardrails for LLM CLIs

Building RAG infrastructure and vector pipelines for large-scale LLM applications. Developing realistic synthetic data generation for enterprise use cases.

• →RAG & vector infrastructure at scale
• →LLM evaluation datasets (18K+ entities)
• →Multi-turn conversational AI
• →Agent ecosystems with MCP protocol

Architected distributed Python workflows for processing 50K+ PDFs monthly. Built sourcing and filtering automation reducing manual work by 70-80%.

• →50K+ PDF processing systems
• →Distributed Python workflows
• →Sourcing/filtering automation
• →70-80% manual reduction

Optimized OCR + CV document processing pipeline achieving 78% to 92% accuracy improvement and cost reduction from $350/mo to $0.

• →OCR + CV pipeline ($350/mo → $0, 78% → 92%)
• →20K+ document processing monthly
• →3.0s → 1.1s latency optimization
• →Production ML system scaling

Developed landslide risk prediction models achieving 90.1% accuracy using Logistic Regression, Linear Regression, CatBoost, and LightGBM on geospatial and environmental datasets.

• →Landslide risk prediction models
• →Logistic Regression, Linear Regression, CatBoost, LightGBM
• →Geospatial & environmental data engineering
• →90.1% accuracy on operational risk analysis