Skip to content
View rhnfzl's full-sized avatar
:octocat:
Focusing
:octocat:
Focusing
17 followers · 17 following

Achievements

Achievement: Pair ExtraordinaireAchievement: StarstruckAchievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pair ExtraordinaireAchievement: StarstruckAchievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLO

Block or report rhnfzl

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
rhnfzl/README.md

Rehan Fazal - ML and NLP Engineer building production NLP systems, semantic search, RAG, and agentic AI

About Me

Hey, I'm Rehan. ML/AI Engineer designing and shipping production machine learning and Natural Language Processing (NLP) systems end-to-end, from training custom language models and embedding pipelines to deploying real-time classification APIs at scale. Architecting across the full stack: transformer fine-tuning, semantic search, RAG systems, agentic AI, LLM orchestration, and ML infrastructure. Currently leading an R&D team building multilingual NLP systems serving English, Dutch, and German languages.

MSc in Computer Science & Data Science · Eindhoven University of Technology (TU/e)

When not training models: runner, boulderer, speed listener, and occasional traveler.

How I Work

  • If something has to be done twice, the third time it's automated into a pipeline. I'd rather spend a day automating than a week doing the same thing manually.

  • Don't believe in guessing thresholds. Measure them from both quantitative and qualitative perspectives. When multiple approaches each get something right, combine them and let the data decide. Everything gets tested before it ships.

  • Build models from scratch and put them in production. If it can't run unattended at scale, it's not done.

What I Build

RAG, Agents & LLM Systems
Agentic AI architectures with tool orchestration and MCP servers auto-generated from API specs. Multi-LLM judge-based evaluation with cross-review. RAG pipelines with ChromaDB, LanceDB, Qdrant, and FAISS. Prompt optimization with structured output.

Architecture & MLOps
Designed AI infrastructure layer over existing SaaS platform: event-driven multi-agent orchestration, MCP tooling, SSE streaming, observability, and EKS deployment. Domain-agnostic shared infrastructure across production classifiers. SageMaker endpoints, Redis caching, MLflow, CI/CD pipelines.

Semantic Classification Systems
Production multilingual classifiers using transformer embeddings. Low-latency inference at scale, deployed as FastAPI services on AWS.

Automation & Developer Tooling
Auto-generated MCP tooling from API specs, autonomous Jira-to-Slack digests, meeting transcript processing pipelines, CI/CD workflows with quality gates.

Custom Language Models
Domain-specific BERT, RoBERTa, and ModernBERT variants trained from scratch with custom tokenizers. Contrastive ranking models for relevance filtering. Deployed on SageMaker GPU endpoints.

Entity Similarity & Embeddings
Multi-algorithm score fusion, embedding aggregation via attention and Set Transformer, empirically calibrated thresholds, LLM-based quality evaluation.

Information Extraction & NLP Pipelines
Document parsing with LLM post-processing, NER-based extraction (GliNER, ONNX ensembles), PII anonymization, multilingual taxonomy enrichment, clustering and outlier detection.

Data & Performance Engineering
Large-scale multilingual data processing with language detection. FAISS index benchmarking, async caching layers, stress testing, and latency optimization for production classifiers.

Tech Stack

Core ML & AI

Python programming language PyTorch deep learning framework TensorFlow machine learning framework NumPy numerical computing Pandas data analysis library scikit-learn machine learning library Hugging Face Transformers ONNX model optimization runtime spaCy NLP library Sentence-Transformers embedding models DeepSpeed distributed training NLTK natural language toolkit

LLM & NLP

OpenAI GPT API Anthropic Claude API LangChain LLM orchestration FAISS vector similarity search ChromaDB vector database LanceDB vector database Qdrant vector search engine Pydantic AI structured output LlamaIndex data framework for LLMs CrewAI multi-agent framework Instructor structured LLM output

Infrastructure & DevOps

Amazon Web Services cloud platform Docker containerization FastAPI Python web framework Linux operating system Redis in-memory data store PostgreSQL relational database MySQL relational database Apache Cassandra distributed database Git version control GitHub Actions CI/CD IBM DB2 database Celery distributed task queue Kubernetes container orchestration Gunicorn Python WSGI server

Tools & Platforms

Streamlit data app framework AWS SageMaker ML platform MLflow experiment tracking Weights and Biases ML experiment tracking Langfuse LLM observability Plotly interactive data visualization Pydantic data validation Ruff Python linter

Currently Learning

Advanced RAG retrieval augmented generation Large Language Model training and fine-tuning Agentic AI autonomous systems Model Context Protocol MCP Rust systems programming language

Exploring advanced RAG architectures, LLM training and fine-tuning, agentic AI, Model Context Protocol (MCP), and Rust for systems programming.

Open Source

SqueakyCleanText Stars reddit-stash-insights Stars reddit-stash Stars

SqueakyCleanText - Text cleaning and NER pipeline for NLP. Multi-backend entity recognition (ONNX, PyTorch, GLiNER), PII anonymization, and multilingual support.

reddit-stash-insights - Semantic search and RAG-powered chat over your Reddit archive. Hybrid retrieval (BGE-M3 embeddings + BM25) with multi-LLM support.

reddit-stash - Automated Reddit backup via GitHub Actions. Archives saved posts, comments, and upvotes to local storage, Dropbox, or S3.

GitHub Stats

Rehan Fazal GitHub contribution statistics

Rehan Fazal GitHub contribution streak

Rehan Fazal GitHub activity graph

Connect With Me

LinkedIn HuggingFace Bluesky X Email

Rehan Fazal GitHub contribution snake animation

Pinned Loading

  1. SqueakyCleanText SqueakyCleanText Public

    Text preprocessing and PII anonymisation for NLP/ML. ONNX NER ensemble, language detection, stopword removal. Built for statistical ML and language models.

    Python 7

  2. reddit-stash reddit-stash Public

    Take the control of your reddit saved post and the comments with context locally or on dropbox

    Python 92 35