> work
Projects
7 projects
Python Logging SDK — Production-Grade Distributed Observability
A lightweight, plug-and-play Python Logging SDK that standardizes structured observability across AI/ML pipelines, distributed systems, and backend services. Automatic context injection (trace_id, request_id, span_id), sensitive data redaction, environment-aware config, and OpenTelemetry-compatible integration with Azure Monitor.
Language-Agnostic API & AI/ML E2E Testing Framework
A plug-and-play reliability engine for modern API and AI/ML systems. JSON-driven DSL, parallel execution, built-in observability with structured logs, traces, and SLA enforcement. Works across Python, .NET, and ML pipelines.
Local MCP Server — Personal DevOps & MLOps AI Assistant
A fully local Model Context Protocol server built with FastAPI and Phi-3 Mini (GPT4All). Understands natural language commands and executes DevOps/cloud tasks — no cloud, no API keys. Hybrid routing: keyword-fast for predictable tasks, LLM for flexible intent.
Local AI Billing Chat
Chat with your cloud billing data in plain language. Upload AWS or Azure CSVs and ask about cost changes, new services, and team breakdowns — fully local using Ollama, no data leaves your machine.
Local-First AI Knowledge Base
A unified local AI chat interface to query PDFs, Word files, PPTs, markdown, and notes entirely on-device. Powered by Gemma 4 (2B) via Ollama, ChromaDB vector search, and NetworkX knowledge graph — with hash-based incremental indexing. Zero cloud dependency.
AI Decision Engine — Mistral, FastAPI & PromptFlow
A dynamic decision engine that evaluates user data against JSON-defined rules and provides AI-driven reasoning using Mistral-7B-OpenOrca via GPT4All. Generates HTML & PDF reports, orchestrated with PromptFlow, deployed on Azure App Services with GPU support and full CI/CD via Bicep.
Local-First AI Security Auditor: Evidence, Not Excuses
A local-first AI-powered security auditor that ingests SARIF/JSON from any tool, deduplicates findings, and outputs a single, evidence-based Markdown report—no hallucinations, no data leakage.