Sindhi NLP
Reproducible Sindhi sentiment analysis with a fine-tuned XLM-RoBERTa model, Python SDK, CLI, and FastAPI service.
01 · @alinawazcode · Sukkur, Sindh, Pakistan
I'm a Python developer who takes machine learning work from a research notebook to something other people can install, call and trust. I build the whole chain: labeled datasets, training and evaluation, then packaging, APIs and deployment.
My NLP projects focus on low-resource languages, which taught me to benchmark carefully and report negative results when the numbers demand it. Workflow: reproducible runs, baselines first, honest reporting.
Fine-tuned XLM-RoBERTa & dual TF-IDF + LinearSVC models, trained tokenizers, and benchmarked 14+ baselines.
FastAPI and Django services, zero-knowledge AES-256-GCM encryption, single-use QR passes, and rate limiting.
10 public GitHub repositories, sindhi-nlp on PyPI, Sindhi Tokenizer v1.0.0, Hugging Face models & Spaces.
Pinned NLP, semantic search, security, and compiler builds across 10 public repositories.
View all 10 repositories on GitHub ↗Reproducible Sindhi sentiment analysis with a fine-tuned XLM-RoBERTa model, Python SDK, CLI, and FastAPI service.
Dual TF-IDF + LinearSVC classifier reaching 91.7% accuracy and 0.918 macro F1 on a human-labeled 4,420-sentence corpus.
Zero-knowledge document sharing with AES-256-GCM encryption, single-use QR passes, rate limiting, and zero plaintext exposure.
SentencePiece Unigram tokenizer trained on 1.46M+ sentences, benchmarked across 17 models with 0.0000% character loss.
Semantic dataset search engine matching queries by meaning with all-MiniLM-L6-v2 embeddings and a Gradio interface.
Custom statically-typed programming language built from scratch in Python with a 9-stage live visual compiler debugger.
01
Core backend engineering, ML pipelines, web scrapers, and interactive web applications.
02
Transformer fine-tuning, subword tokenization, classical baselines, and semantic vector retrieval.
03
Production REST APIs, encrypted web platforms, live ML Spaces, and desktop compiler tools.
04
Reproducible CI/CD automation, containerized services, Python packages, and cloud hosting.