Skip to content
View rsalehin's full-sized avatar
🏠
Working from home
🏠
Working from home

Highlights

  • Pro

Block or report rsalehin

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 is supported. This note will only be visible to you.
Report abuse

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

Report abuse
rsalehin/README.md

Rafiqus Salehin

ML Engineer · Research Software Developer · Physics MSc

I build ML systems at the intersection of scientific computing and production engineering — retrieval pipelines, LLM applications, and simulation tooling for research and industry.

Currently a Research Assistant at Karl-Schwarzschild-Observatorium Tautenburg, working on intracluster magnetic field modelling and AI tooling for radio astronomy.


Selected Projects

Local hybrid RAG and knowledge graph over 1,000+ astrophysics papers. Dense retrieval (SPECTER2) + BM25 with RRF fusion, DuckDB structured metadata, NetworkX citation graph, FastAPI backend, D3.js frontend, Chrome extension.

Python FAISS DuckDB NetworkX SPECTER2 BGE-M3 BM25 FastAPI

ML4SCI Google Summer of Code 2026 — three proposals for gravitational lens finding, unsupervised super-resolution of lensing images, and LSST data processing pipeline. SO(2)-equivariant CNN for lens classification · CycleGAN-style unpaired SR · Butler/LSST pipeline simulation.

PyTorch Deep Learning Astronomy Computer Vision GSoC

CLI tool for converting GitHub repositories and documentation sites into single Markdown files for RAG ingestion, with six enrichment passes including API index extraction and cross-reference tables.

Python CLI RAG Developer Tools LLM

Exoplanet research assistant combining Elasticsearch hybrid search with Vertex AI/Gemini. Built for Google Cloud Hackathon.

Python Elasticsearch GCP Gemini FastAPI Hybrid Search


Production Experience

  • BERT classifier — automated routing of 6,000 daily support tickets at 93.7% accuracy, saving 60 person-hours/week
  • LLaMA 2 13B fine-tuning — LoRA/PEFT on proprietary knowledge base, deployed via vLLM; cut analyst research time by 50%
  • Fraud detection tooling — reduced fraudulent listings by 43% on a classified ads marketplace serving 100k+ users
  • Multi-agent LLM pipeline — LangGraph + DeepSeek with tool-calling for automated astronomical observation report analysis

Stack

ML / AI      PyTorch · Hugging Face · FAISS · LangGraph · vLLM · LoRA/PEFT · SPECTER2 · BGE-M3
Backend      FastAPI · Flask · Node.js · PostgreSQL · DuckDB · MongoDB · Redis
Cloud        AWS · GCP · Docker · CI/CD · Linux
Scientific   NumPy · SciPy · Matplotlib · PyMuPDF · NetworkX · LaTeX
Languages    Python · C++ · Java · JavaScript · SQL · Bash

Background

  • 🎓 M.Sc. Physics — Leipzig University (in progress)
  • 🎓 B.Sc. Physics — Leipzig University (2.1)
  • 🎓 B.Sc. Computer Science & Engineering — BUET
  • 🏆 5th Place, ACM-ICPC Asia Regional
  • 📍 Leipzig, Germany · Open to opportunities in Germany and EU

Contact

LinkedIn Email Portfolio

Pinned Loading

  1. ai-projects ai-projects Public

    Jupyter Notebook

  2. CodeGuardian CodeGuardian Public

    A Hackathon Project develop for AWS Global AI Agent Hackathon

    Python 2

  3. project-kepler-elastic project-kepler-elastic Public

    Python 1

  4. DeepLense-GSoC-2026 DeepLense-GSoC-2026 Public

    Tests and Results for Proposals

    Jupyter Notebook 1

  5. icm-magnetic-field-kb icm-magnetic-field-kb Public

    Local hybrid knowledge graph over ICM magnetic field literature

    Python 1