File size: 3,719 Bytes
01316f7 2d72b32 01316f7 2d72b32 01316f7 2d72b32 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 | ---
title: Alexandru Cojocari — Python Backend & GenAI Portfolio
emoji: 🧩
colorFrom: blue
colorTo: purple
sdk: static
app_file: index.html
---
# Alexandru Cojocari
**Python Backend Engineer · GenAI & Applied AI · QA Automation**
I have around 10 years of professional Python backend experience and commercial Java experience. My portfolio explores reliable backend systems and practical AI engineering, with QA automation as a secondary strength.
I build small, runnable demos around the engineering details that matter: safe interfaces, reproducible evaluation, data integrity, and useful tests. These are portfolio projects, not claims of production deployment.
[GitHub profile](https://github.com/DatadrivenAlex)
## Selected projects
### AI and GenAI engineering
- [Model Rollout Gate](https://github.com/DatadrivenAlex/model-rollout-gate) — Offline champion–challenger checks for binary models: paired bootstrap intervals, segment gates, an empirical p95 latency budget, and a replayable JSON decision. Uses synthetic data; it informs a human release decision and does not deploy models.
- [LLM Gateway Workbench](https://github.com/DatadrivenAlex/llm-gateway-workbench) — FastAPI demo of request idempotency, per-tenant quotas, provider adapters, bounded retries and fallback. The included provider is deterministic/offline; this is not a production gateway.
- [Document AI Extractor](https://github.com/DatadrivenAlex/document-ai-extractor) — Extracts invoice and receipt fields from PDFs/images with pdfplumber, Tesseract OCR, rules, optional provider fallback, and Pydantic financial checks. Scoring uses a synthetic corpus; the optional live API is not exercised in CI.
- [AI Security Guardrails](https://github.com/DatadrivenAlex/ai-security-guardrails) — Rule-based HTTP firewall that can allow, deny, or rewrite text before or after an LLM call. Offline and deterministic; it makes no model calls and is not a production WAF.
- [NL2SQL Analytics Agent](https://github.com/DatadrivenAlex/nl2sql-analytics-agent) — Read-only SQL analytics over a small SQLite demo warehouse, with SQLGlot parsing and SQLite authorizer safeguards. The default path uses a fixed offline question catalog; an OpenAI-compatible endpoint is optional. Not a production text-to-SQL system.
- [Semantic Search & Embeddings](https://github.com/DatadrivenAlex/semantic-search-embeddings) — Compares BM25, FAISS dense retrieval, and reciprocal-rank fusion on a project-authored FAQ corpus. The checked-in evaluation uses a deterministic hashing embedder; CI does not run the optional neural model.
### Backend, data and quality
- [ML Dataset Lineage Registry](https://github.com/DatadrivenAlex/ml-dataset-lineage-registry) — FastAPI service for schema-validated, immutable dataset snapshots with canonical SHA-256 fingerprints and parent lineage. Uses SQLite and synthetic data; it has no authentication or tenant isolation.
- [Playwright QA Automation](https://github.com/DatadrivenAlex/playwright-qa-automation) — Self-contained Python test framework for a demo shop: httpx API tests, Playwright UI tests, page objects, pytest, and failure artifacts. It does not contact an external store or perform payments.
## Engineering focus
Python · FastAPI · Java · SQL · API design · GenAI integration · retrieval and evaluation · data validation · model rollout checks · security guardrails · pytest · Playwright · CI
## About the demos
The repositories are intended for learning, review, and portfolio discussion. Each README documents its own assumptions and limits. Unless a project says otherwise, examples use synthetic or local demo data; do not send confidential data to these demos.
|