I build AI systems, developer tools, and energy-aware infrastructure.
The thread through my work is measurement: LLM agents should be evaluated, production systems should be observable, and energy claims should be reproducible before they are optimized.
My current focus is the overlap between:
- LLM systems: tool-calling agents, retrieval, structured outputs, and evaluation loops.
- Backend systems: APIs, software architecture, automation, and production-facing developer tools.
- Energy-aware software: measurement, benchmarking, profiling overhead, reproducibility, and regression detection.
- Research prototypes that become usable tools, not only demos.
Wattch is lightweight energy-profiling infrastructure for developers and AI coding agents.
It is a Rust daemon and CLI for low-overhead energy measurement, built around reproducible benchmarks and machine-readable reports. The long-term direction is IDE and MCP integration, so AI agents can detect energy regressions, reason about measurement confidence, and suggest greener code changes without hiding overhead or uncertainty.
Status: work in progress.
- jreferral - recommends energy-efficient JVM configurations for Java software.
- IJoules - measures energy consumption of Python code on macOS / Intel CPU.
- Wattch - current project; Rust daemon and CLI for developer-facing energy profiling.
My PhD work focused on energy-aware software engineering: measurement, benchmarking, testing, optimization, language/runtime behavior, and reproducibility. Wattch builds on that foundation by turning the research concerns into developer infrastructure: repeatable runs, explicit overhead, structured reports, and tooling that can fit into an engineering workflow.
- chakib_belgaid_thesis - thesis source and materials.
Source: thesisBrainMap.svg · thesisBrainMap.drawio
- Measurable before optimized.
- Local-first when possible.
- Explicit about uncertainty and overhead.
- Useful to developers, not only impressive in demos.