Rabih Jabr

A retrieval system over a large internal Confluence wiki. Engineering teams can ask questions against their own documentation and the source code it describes. It refreshes on a schedule instead of indexing live, which is what keeps queries fast and inference cost low.

• A weekly pipeline exports a Confluence tree to Markdown starting from a configurable root node, so any team can point the system at its own domain without a code change.
• A scoring stage ranks every page before it reaches the index. It weighs concept relevance, how well the page is maintained, whether it has been deprecated, and whether it links to or mentions real source code. Stale pages never get far enough in to dilute retrieval.
• Resolves the source-code half of that relevance signal against both Perforce and Stash (Bitbucket).
• Per-team pipelines provision and refresh each team's index independently.
• Runs on locally hosted Llama models as well as hosted APIs. I benchmarked across model choices to confirm the cheaper local setup held up on retrieval quality.

An event discovery platform I'm designing and building on my own, end to end: Next.js App Router on the front, a Spring Boot REST API behind it, PostgreSQL underneath, and JWT auth.

• Built on Next.js 16 App Router with React 19 Server Components, backed by a Spring Boot REST API and JPA persistence.
• Designed the PostgreSQL schema for the calendar, bookmarking, messaging, and notification systems.
• Implemented an interactive infinite-scroll feed with CDN-backed image optimization and public-URL sharing to drive organic growth.
• Built with heavy AI assistance (Claude Code). That is the reason one person can cover schema design through to a production frontend.

A production desktop application I designed, built, and delivered single-handedly for a manufacturing client. Runs fully offline on a single machine, handling customer intake, production yield calculation, invoicing, inventory, and financial reporting.

• Designed the relational schema and migration strategy on SQLite, then covered the money, allocation, and stock logic with unit tests, including regression tests for bugs that had already shipped.
• Built machine-bound license verification (HMAC) and an automated backup pipeline for a single-operator environment with no network and no second pair of eyes.
• Delivered a bilingual interface (Arabic RTL / English LTR) with custom HTML-to-PDF printing for invoices and financial reports.
• Maintain the codebase's documentation as a structured, agent-readable knowledge base of domain rules and invariants, so AI agents make correct changes without re-deriving the business logic each session.