I built a privacy-first AI assistant that automatically resolves IT errors by extracting information from screenshots, searching knowledge bases, and generating solutions—while protecting sensitive client data.

Manual troubleshooting delays: IT support teams spent hours searching through documentation for error solutions

Screenshot-based tickets: Users submitted error screenshots that required manual text extraction and analysis

Privacy risks: Existing tools sent sensitive client data to external APIs, violating compliance requirements

Knowledge fragmentation: Error solutions scattered across multiple PDFs, databases, and documentation sources

Escalation bottlenecks: No automated way to draft escalation tickets when solutions weren't found

Developed ARES (Automated Resolution & Escalation System) using LangGraph workflow orchestration with:

Vision LLM OCR: Ollama-based text extraction from error screenshots

Dual RAG System: Separate vector databases for errors (Oracle, MySQL, MITRE) and general knowledge

Smart PII Detection: Pattern-based filtering that distinguishes error codes from sensitive data

Interactive Workflow: LangGraph interrupts for user approval with automatic escalation on rejection

Advanced Document Processing: Docling v2 for PDFs with table/chart recognition

Tech Stack: Python, FastAPI, LangGraph, ChromaDB, Ollama (gemma2:4b), Qwen embeddings, Streamlit

Privacy-First: Rejects requests containing client information automatically

Multi-Format Support: Processes PDFs, images, CSV, JSON, Markdown, HTML

GPU Acceleration: CUDA/MPS support for faster embeddings and inference

Stateful Workflows: Checkpointed execution with interrupt/resume capability

Comprehensive Testing: 80% code coverage with pytest suite

Production Ready: Structured logging, metrics tracking, error handling

6-node workflow with conditional routing and human-in-the-loop approval

Dual knowledge bases indexed from 13 documents (6 error docs + 7 knowledge docs)

Real-time PII detection with 95% confidence scoring

Automated escalation drafting when solutions aren't found

25 test images validated for OCR accuracy

I specialize in developing production-grade AI agents with LangGraph, RAG systems, and workflow orchestration. Whether you need document processing, intelligent automation, or privacy-focused AI solutions, I deliver scalable systems that work.

Let's discuss your project → Send me an message