This work comes from more than a decade of owning legacy systems from the inside: as a senior engineer and technical lead responsible for systems that couldn't go down, and more recently as an independent consultant. Details are anonymized to protect confidentiality; references are available on request.
1
Media Analytics Company
70%faster reports
45s → 10sgeneration time
Biweekly → Dailydeploy frequency
PHPCI/CDDatabase OptimizationAutomated Testing
The Situation
A media analytics company had a PHP application that powered their core reporting platform. The system had grown organically over 8 years and was showing its age.
The application was experiencing frequent timeouts during report generation, deployments were manual and error-prone, and the development team was afraid to make changes due to lack of test coverage.
What I Did
Conducted a full system audit to identify root causes
Implemented database query optimizations that reduced report generation time by 70%
Established a CI/CD pipeline with automated testing
Introduced progressive refactoring patterns the team could continue
Outcome
Report generation dropped from 45 seconds to under 10 seconds. Deployment frequency increased from biweekly to daily. The team gained confidence to make changes without fear of breaking production.
2
Advertising Data Analytics Company
Featured
6MCP tools built
FullOAuth 2.1 server
Cross-VPCarchitecture
MCPOAuth 2.1AWS LambdaAPI GatewayPostgreSQLIaC
The Situation
A national advertising data analytics company maintained a large database of local market advertising spending data used by media companies and agencies across the US. Their clients accessed this data through a traditional web dashboard, but the company wanted to make it available where users already work: AI assistants, CRMs, internal tools, and other integrated platforms.
The company's data lived in a private database behind a VPC with a complex permissions model: users belonged to companies with product subscriptions that granted access to specific markets and tools. There was no API layer suitable for external consumption, and exposing the data required production-grade authentication that connected clients could negotiate automatically.
What I Did
Designed and built a remote MCP server that exposes advertising data to connected clients over authenticated HTTPS using the Model Context Protocol
Implemented a full OAuth 2.1 authorization server with dynamic client registration, PKCE, JWT-based access tokens, and magic link authentication, integrated against the existing user database
Built 6 MCP tools (market lookup, spending breakdowns, trend analysis, category rankings, media mix, market comparison) with per-request permission enforcement
Deployed on AWS with Lambda, API Gateway, VPC networking, and infrastructure-as-code, including cross-database architecture spanning existing production data and new MCP-specific tables
Outcome
Authorized users can now access advertising spending data through any MCP-compatible client, whether that's an AI assistant like Claude, an IDE like Cursor, or integrated directly into their existing business tools. The system enforces the same permission model as the existing platform: users only see data their subscription allows. The server handles the full OAuth handshake automatically, so connecting is as simple as entering a URL.
3
E-commerce Platform
200+businesses served
Globalmarket reach
Multi-stackcodebase unified
Multi-stackCI/CDGlobal MarketsIntegration Testing
The Situation
A large-scale e-commerce platform served hundreds of dealer businesses across global markets. The system had grown over several years with multiple technology stacks and was becoming difficult to maintain.
Critical integrations between systems had undocumented behavior. Deployment processes were fragile. The codebase spanned multiple languages and frameworks, making it hard for any single engineer to work across the full system.
What I Did
Developed and maintained client-facing systems supporting 200+ businesses across global markets
Documented undocumented integrations through code analysis and testing
Designed CI pipelines enabling frequent, reliable deployments
Established patterns for working across the multi-stack codebase
Outcome
Deployment reliability improved significantly. The team gained the ability to ship changes confidently across the platform, and institutional knowledge was captured in documentation rather than locked in individual engineers' heads.
A lean B2B analytics firm had outgrown their initial architecture. What worked for 100 customers was failing at 1,000.
Database queries that were fast with small data sets were now causing timeouts. Background job processing was unreliable. The team was spending more time firefighting than building features.
What I Did
Identified the top 10 problematic queries and optimized them
Redesigned the background job architecture for reliability
Implemented proper monitoring and alerting
Created runbooks for common operational issues
Outcome
System stability improved dramatically - from multiple incidents per week to less than one per month. The team shifted from reactive firefighting to proactive feature development.
Have a Similar Challenge?
Let's discuss how I can help with your legacy system.