Mohamed Hassan

• Lead and mentor a pod of three engineers that owns AIMY QA and Connect: sprint planning, estimation, stakeholder triage and code review (160+ pull requests reviewed since July 2026). The pod has delivered 280+ tickets at about 25 a sprint, with 5% reopened.
• Designed an AI actions layer and Product Insights. The model classifies interactions into facts, SQL ranks them into insights, and a person approves every action on a six-state machine with an audit trail and undo. The model reads and explains; it never changes state.
• Built the Ask AiMY analytics agent (FastAPI, Claude through LiteLLM): full ticket context, answers scoped to one interaction, the same numbers as the dashboard, and a guard that stops it from inventing data it cannot see.
• Built MCP servers that connect Claude and in-house agents to knowledge retrieval, sales enrichment and recruitment workflows.
• Built config-driven knowledge sync for Freshdesk and TeamSupport, added context-tag reranking in Qdrant, and summarised tickets over the 8,000-token embedding limit instead of failing them.
• Designed the central ABAC / RBAC authorization system used by 1,000+ employees across 6+ applications, with matching .NET and Python middleware; later moved access grants to entity keys across 4 services and closed a cross-organization scoping gap.
• Unified voice-call persistence across Voice, Connect and QA: one post-call path that stores every transcript, recordings through presigned S3 URLs, and race-safe session writes enforced by unique indexes.
• Led the AIMY QA 2.0 release across 6 repositories and coordinated platform releases of up to 14 repositories, plus 20+ production hotfixes; wrote the team's branching and release workflow.
• Wrote WarrenMQ, a RabbitMQ library published on NuGet, and a Mediator library that replaced MediatR after its 2025 licence change; both are reused across services. Also built 9 production n8n workflows that keep Salesforce, Zendesk, Freshdesk and Jira in sync, and moved 8+ containerised applications between AWS accounts and regions.

• Rebuilt a legacy codebase of more than 20 projects into 4 Clean Architecture projects (CQRS, DDD, no circular references), then grew it to 240+ REST endpoints and 260+ MediatR handlers serving 6 roles, fully bilingual in English and Arabic with right-to-left layout.
• Raised the concurrent-user ceiling from about 50 to 2,000+ and proved it under load. I removed N+1 queries, moved 65 read handlers to Dapper and added HybridCache with stampede protection. A 12-scenario k6 suite then held 2,000 concurrent users with server p95 flat at about 34 ms on one replica, and control experiments traced the remaining limits to configuration, not code.
• Per-organisation single sign-on (OIDC with just-in-time provisioning), branded sign-in pages and emails for each organisation, and refresh tokens, on fail-closed defaults: rate limiting, account lockout, Key Vault-rotated JWT keys and PII masking.
• Real-time delivery alerts: SignalR order tracking that resyncs after connection gaps, and Web Push (VAPID, service worker, installable PWA including iPhone) so delivery staff get new-order and VIP alerts with the app closed.
• Inventory and recipe engine: receiving with batch, expiry and unit price; adjustments, waste and stock counts; recipes that deduct stock on every order, unit conversions by kind, and how many orders current stock covers.
• Tenant isolation and concurrency: store-level scoping on every inventory, recipe and unit action; vendor-operated stores serving several organisations behind approval-gated linking; atomic numbering with UPDATE … OUTPUT, row-version concurrency on orders, and a filtered unique index for one active cart per user.
• Azure end to end: moved the platform from a legacy Azure tenant to the client's own directory and DevOps organisation, with separate production (UAE North) and test environments; a build-once pipeline with manual production approval and a health gate; media moved from SQL to Cloudflare R2 with presigned uploads; about 20% lower cost with scale-to-zero.
• Testing: grew the Testcontainers SQL Server suite from 561 to 1,260+ tests, with a custom interceptor that fails CI when a change adds N+1 queries.

240+ API endpoints260+ handlers51 data entities1,260+ tests140+ Angular components560+ commits

• Redesigned roles and permissions: more than 15 actions per resource in place of view and manage, and a new resource is now added in one place with no database migration.
• Cut one list's response time from 2 minutes to 235 ms.
• Built, over seven months, the feature that opened the system to outside vendors alongside employees, each with their own interface; rewrote the groups and rules services across 70+ files and fixed 40 validation bugs.
• Integrated Azure Active Directory sign-in, and built a templating system for email, SMS and notifications that cut content-update effort by 40%.
• In one year: 240+ work items and 500+ story points completed, 30+ pull requests written and 60+ reviewed.
• ServiceDesk Plus portals for a bank, a government authority, a healthcare procurement company, a cloud provider and a gaming group: employee dashboards on the product's API in plain JavaScript, an Arabic right-to-left service portal, and branded login pages. The product has no extension point for most of this, so I traced how its pages load and changed the bundled scripts and templates, keeping every original so each change could be undone.