Projects Showcase
A few systems from recent work. Some were built with large teams. A few are personal projects, done outside the job.
Featured
GitHub Agentic SDLC
A proof of concept for the whole software delivery lifecycle: a user story from any work tracker, such as Jira or Azure DevOps, goes through planning, code, tests, review, merge and release, and the story is closed at the end. The work is done by a team of six GitHub Copilot agents: one orchestrator that calls five specialist sub-agents only when it needs them. A separate pull-request review agent then checks the result in its own workflow, on a different model from the one that wrote the code. People still approve the plan and merge the code. It is built on GitHub Agentic Workflows (gh-aw), GitHub's way of running coding agents inside GitHub Actions: each workflow is a Markdown file with plain-language instructions for the agent and a short settings block for triggers, permissions and allowed outputs, which the gh aw compile command turns into a standard Actions workflow. The agents can read but not write; anything they want to publish is checked and carried out by a separate job. There are no stored passwords or tokens, and every run leaves logs that can go to Azure Monitor, Splunk, Datadog, or any other analytics tool.
Hybrid RAG Knowledge Platform
A proof of concept for a retrieval platform, built to see how the pieces behave together before relying on them. A .NET 10 service collects Stack Overflow questions through a Redis queue, splits them into chunks, and indexes them in PostgreSQL with pgvector. An agent built on Microsoft Agent Framework answers from that corpus using hybrid search, vector and full-text results merged with Reciprocal Rank Fusion, and links each answer to its sources. Model calls go through a LiteLLM gateway, so changing provider, for example from OpenRouter to Azure OpenAI, is a configuration change rather than a code change. It runs in Docker Compose, with AKS and KEDA manifests drafted as the next step.
RentGuide AI Concierge
Guests arrive at an unfamiliar property with the same questions: how the thermostat works, where to park, what time checkout is, what is worth eating nearby. RentGuide answers in plain language, grounded in the property's own documents so it does not invent details, and uses images when a photo is clearer than text. Currently in pilot.
Agentic AI Orchestration
Agentic AI for enterprise workflows, with the governance written down: multi-agent collaboration through Agent-to-Agent (A2A) communication, Chain of Thought and Chain of Debate reasoning, a Shared AI Plugins Marketplace for engineering teams, pull-request review grounded in internal engineering rules, and Model Context Protocol (MCP) tooling with Entra ID authorization.
More projects
NutriScan AI – Multimodal Nutrition Scanner
A phone photo of a meal, and an estimate of its nutrition. Built on Gemini, with a step to correct the estimate when it is wrong. Three parts: a React and Vite web app, a Telegram bot for when a website is inconvenient, and a containerized Node and Express API. Hosted on Azure and deployed with GitHub Actions. A personal project.
Gemini AI multimodal vision
Ovinojo AI – Multimodal Wine Sommelier
A photo of a wine label, and a short reading: body, tannins, acidity, sweetness, alcohol, what to serve it with, and at what temperature. Search grounding is used so the answer is not limited to a fixed list of bottles. It follows NutriScan: a React 19 app, a Telegram bot, and a containerized API with Swagger, on Azure, with automated deploys. Reusing that shape made the second one much smaller.
Gemini multimodal wine recognition
Intelligent Telegram Agent
A Telegram bot that asks what it still needs to know, instead of acting on a guess. The extra question matters once a choice depends on more than one or two details.
Conversational clarification loop
Enterprise Azure Cloud Architecture & Modernization
Enterprise architecture direction across the IT portfolio, 30+ concurrent projects and initiatives from early outlines and intake through delivery, supporting 8+ delivery teams on Azure. Production in West US 3 runs across three availability zones. End-of-life NGINX ingress was replaced with Istio Gateway API and a sidecarless ambient service mesh, ahead of the November 2026 support cutoff. Zero-trust identity covers employees, partners, and customers (Entra ID, Entra B2B, Azure AD B2C / Entra External ID, passwordless Microsoft Authenticator), federated workload credentials with no stored secrets, and privileged access management with CyberArk PAM and Entra PIM. A shared CI/CD orchestrator on GitHub Actions provides versioned, reusable workflows with build, security-scanning, and deployment gates. Traced and eliminated a Log Analytics cost leak, cutting log-ingestion costs by about 90%. Microsoft Fabric is connected privately. Architecture decisions go through an Architecture Review Board (ARB), using RFC 2119 standards.
~90% cut in log-ingestion costs
Enterprise Platform Migration to AWS
25+ enterprise applications moved from on-premises infrastructure to AWS. They were redesigned as they moved, rather than copied as they were. Customer and employee identity used Duende IdentityServer for borrower portals over OIDC/OAuth 2.0/JWT with MFA, and Ping Identity federated with Azure AD / Entra ID for internal SSO. Operational cost came down by about 25%, and scaling became something the team could plan, rather than something that happened during an incident.
25+ apps migrated
Real-Time Data Workflow Application
A single-page app and web API that give customers a live view of their data and the workflows running against it, in place of a report that depended on exported spreadsheets and the previous day's numbers.
Real-time visualization