Software Engineering student at Sichuan University, China, graduating in June 2027.
My current academic focus is AI for Software Engineering, especially LLM-based program repair, coding-agent reliability, and empirical evaluation. I also build systems projects to strengthen my understanding of distributed systems, fault tolerance, persistence, and software reliability.
I am currently preparing for Fall 2027 Master's applications.
Completed independent empirical Software Engineering study; manuscript in preparation.
I studied whether a coding agent can accurately judge, before attempting a repair:
- how likely it is to successfully fix a repository-level software bug, and
- whether the repository evidence it has been given is sufficient.
The final study used:
- 60 real repository-level software bugs
- 3 controlled evidence conditions
- 3 replicates per task-condition
- 540 repair attempts
- executable Docker-based evaluation
- a frozen experimental protocol and analysis plan
- 10,000 task-level bootstrap resamples
Relevant repository evidence substantially improved both confidence and actual repair performance:
- predicted repair success increased by 55.1 percentage points
- executable repair success increased by 20.0 percentage points
However, task-level confidence changes showed very little correspondence with task-level performance improvements (Pearson r = 0.073).
This suggests that coding models can recognize that relevant repository evidence is generally useful while still having limited ability to estimate how much that evidence will help on a specific software-repair task.
A public research artifact will be added after the manuscript and replication package are prepared.
A coordinator/worker distributed task-processing system built from scratch in Java 21.
Engineering highlights
- Custom length-prefixed JSON protocol over TCP
- Single-writer coordinator event loop
- Concurrent worker execution
- Heartbeat-based worker failure detection
- At-least-once execution semantics
- Attempt leases and stale-result rejection
- Automatic retry and reassignment
- SQLite-backed durable state and coordinator restart recovery
- Task deadlines and cooperative cancellation
- Admission control and overload handling
- Durable idempotent submissions
- Persistent priority scheduling
- Deterministic FIFO sequencing
- Aging-based starvation prevention
- GitHub Actions CI
- 172 automated tests
- 119 unit tests
- 53 integration tests
- 0 failures
Technologies: Java 21 · TCP · Concurrency · SQLite · Maven · JUnit · GitHub Actions
Casablanca-focused activity-discovery and outing-planning product built around structured, venue-verified data.
I lead product direction, venue-data design, QA, and an AI-assisted development workflow.
Current scope
- structured dataset covering 99 verified venues
- personalized outing-planning flows
- activity discovery and filtering
- favorites, collections, comparison, and saved outings
- bilingual English/French support
- venue verification and data-quality workflow
- branch-and-PR development process
Technologies: React · TypeScript · Express · Supabase · PostgreSQL · Vite · GitHub Actions
Originally developed as a company training project and later independently extended.
Work includes
- Hadoop/HDFS distributed data processing
- Hive/HiveQL analytics
- Python and NLTK review-text analysis
- Flask Text-to-SQL interface
- schema-aware LLM prompting
- SELECT-only SQL validation
- automatic retry for failed queries
- NOAA weather-data enrichment
Technologies: Python · Hadoop · HDFS · Hive · SQL · NLTK · Flask
Programming: Python · Java · JavaScript · TypeScript · C · C++ · SQL
Systems & Data: TCP sockets · Concurrency · SQLite · Maven · JUnit · Docker · Linux · Hadoop · HDFS · Hive
Web & Tools: React · Express · Flask · Supabase · Git · GitHub Actions
Sichuan University
Bachelor of Software Engineering
English-taught program
Expected graduation: June 2027
Arabic — Native
French — Fluent, DELF B2
English — Fluent, English-medium degree
Mandarin Chinese — Intermediate
- Email: achraf2026@163.com
- GitHub: github.com/achraf059
- Location: Chengdu, China