Backend engineering · API design
Idempotency keys: making a FastAPI intake endpoint safe to retry
Client-generated keys, request-hash checks, atomic writes, and expiration so a retried request never becomes a duplicate row.
Technical writing
Notes on parsing, private AI, and computer science learning
Practical guides grounded in Yusuf Gadelrab’s work in freight intake parsing, CS tutoring, and bilingual computer science education.
Backend, APIs & data engineering
AI systems · Backend engineering
Confidence gating for LLM-extracted fields: deciding what a human must review
Per-field confidence tiers, evidence-span matching, and deterministic cross-checks so review time goes to the fields that actually need it.
Backend engineering · Testing
API integration tests: prove the contract without calling production
Use representative fixtures, strict outgoing-request assertions, deterministic clocks, and a narrow live smoke check to make API changes reviewable.
Backend engineering
Integrating third-party APIs: the failure modes that only appear in production
Timeouts, retries with jitter, idempotency keys, rate limits, pagination, and webhook verification.
Backend engineering · Observability
Structured logging in Python: turning print statements into a debugging tool
JSON formatters, LoggerAdapter for shared context, and a level discipline that keeps alerts meaningful instead of noisy.
Testing · Python
Pytest fixtures and parametrize for a data pipeline: testing the same step against many inputs
Fixtures for shared setup, parametrize for input variation, and when a parametrized fixture is worth the extra layer.
Data analytics · Python
A pandas data-cleaning pipeline that stays debuggable
Named single-purpose steps, row-count logging between them, and an explicit validation pass instead of one dense method chain.
Databases
SQL joins and indexes: the parts that quietly change your answer
Why a LEFT JOIN plus a WHERE becomes an INNER JOIN, how NULLs break NOT IN, and when an index is ignored.
Backend · APIs
Cursor versus offset pagination in a REST API
Why OFFSET duplicates and skips rows on a table still being written to, why deep pages get slower, and how a keyset cursor with a unique tiebreaker fixes both.
Databases
Alembic migrations you can actually roll back
Why autogenerate turns a renamed column into data loss, how to write data migrations that keep working, and the three-deploy pattern for adding a NOT NULL column.
Freight intake automation
Designing a freight intake schema for messages that were never forms
How to preserve evidence, represent missing information, and build a reviewable intake workflow.
Existing guide
Parsing unstructured freight messages into structured load data
How to combine normalization, deterministic patterns, constrained models, and human review.
Developer tooling, build systems & open source
Open source
Writing a secret-detection rule for gitleaks: what a pattern has to prove
How the Supabase token shape was established from evidence, why one key is deliberately not flagged, and what the false-positive tests pin down. The pull request is open and unmerged.
Build tooling
SOURCE_DATE_EPOCH: why a build timestamp is a reproducibility bug
Deterministic SBOM timestamps, defining what counts as a valid value, and the test-isolation trap that appears the moment you read an environment variable.
Developer tooling
Docker for CS coursework: reproducible environments without falling down the rabbit hole
When a container is worth it, a minimal Python and C++ setup, bind mounts, and the mistakes that cost an evening.
Local-first document parsing
Run small language models where the data lives
A practical architecture for extracting useful fields from private client text without making a hosted model the default data path.
CS education & bilingual learning
Teaching
Tutoring computer science when the student is also learning English
Telling a language gap from a concept gap in ninety seconds, keywords as symbols, reading error messages by shape, and what the research does and does not support.
Conference notes
Notes on a SIGCSE TS 2026 poster: what two hours at a poster board actually demands
What the poster format can and cannot carry, using the bilingual coding poster Yusuf co-authored as the worked example.
Bilingual CS education
Bilingual computer science education starts by separating logic from language
A field guide for language-aware instruction, grounded in Yusuf’s ACM SIGCSE 2026 poster collaboration.
CS tutoring
A computer science tutoring guide for learning to explain, trace, and debug
Concrete study and debugging routines drawn from Yusuf’s tutoring work at SJSU since August 2024.
Bilingual programming
Programming syntax can be the barrier—not programming logic
A focused introduction to separating conceptual understanding from English-heavy programming notation.
Programming learning
Learning programming across languages without treating syntax as the whole skill
A method for identifying transferable concepts, comparing syntax, and testing assumptions with small experiments.
Academic recovery
Academic recovery in computer science: a practical reset
A candid, employer-safe guide to assessing a difficult term and building a sustainable recovery routine.
Search & SEO
Search and SEO
The future of search engines: what AI changes about SEO, and what it does not
Zero-click data, retrieval versus ranking, structured data and entity identity, the llms.txt myth, and the crawler economics that will decide what the corpus contains.
More writing
Related work
Research context and poster citation
Explore the ACM SIGCSE 2026 poster collaboration and Yusuf’s 2025 SJSU research presentation.