Muzzamil Rasully

ML & AI Engineer building NLP, RAG and applied machine-learning systems

I build practical ML and AI projects using Python, PyTorch, scikit-learn, Transformers and LangChain — from model training and evaluation to retrieval pipelines, APIs and demos.

View Projects

Open to ML Engineer, AI Engineer, NLP Engineer, RAG/LLM Engineer and ML-focused Data Scientist roles.

rag_pipeline.pyPython
from transformers import AutoModel
import torch

# retrieve, then generate a grounded answer
ctx = retriever.search(query, k=5)
prompt = build_prompt(query, ctx)
answer = llm.generate(prompt)
# → grounded, cited, on-device
10+
ML / DL models trained
3
end-to-end AI systems
0.82
ROC-AUC · churn classifier

Currently focused on

Applied MLRetrieval systems (RAG)Model evaluationMLOps & deploymentData analytics
Open to
ML EngineerAI EngineerNLP EngineerML-focused Data Scientist
Best proof
NLP Retrieval & RAGFinBERT stock ML pipeline0.82 ROC-AUC churn modelCNN from scratch
Core stack
PythonPyTorchscikit-learnTransformersLangChainFAISSFastAPIDockerSQL
Projects

Featured Projects

Practical ML and AI projects — clear problem, method and result. GitHub and case studies available.

NLP Retrieval & RAG

Builds a question-answering system over documents using retrieval, re-ranking and generated answers.

  • Built: TF-IDF, dense and hybrid retrieval baselines
  • Added: FAISS search, cross-encoder re-ranking and Flan-T5 answer generation
  • Proof: GitHub and case study
  • Python
  • FAISS
  • Transformers
  • Flan-T5
  • NLP
  • RAG
Case StudyGitHub

FinSight AI

Tests whether short-term stock movement can be predicted using market data, sentiment and document context.

  • Built: leakage-safe financial ML pipeline with technical indicators and FinBERT sentiment
  • Finding: short-horizon prediction was only slightly above baseline, showing honest evaluation
  • Proof: reproducible pipeline, dashboard, Docker and case study
  • Python
  • PyTorch
  • FinBERT
  • Transformers
  • RAG
  • scikit-learn
Case StudyRepo available on request

Telco Churn Predictor

Predicts customer churn and explains the drivers behind each prediction.

  • Built: two-stage ML pipeline with GBM risk scoring and XGBoost classification
  • Result: 0.82 ROC-AUC on the IBM Telco dataset
  • Explainability: SHAP used to surface churn drivers
  • Python
  • XGBoost
  • scikit-learn
  • pandas
  • SHAP
Case StudyGitHub
Skills

Skills

ML / Modelling

  • Python
  • PyTorch
  • scikit-learn
  • XGBoost
  • Model evaluation

NLP / RAG

  • Transformers
  • LangChain
  • FAISS
  • Flan-T5
  • FinBERT
  • Embeddings

Deployment

  • FastAPI
  • Streamlit
  • Docker
  • MLflow
  • GitHub

Data foundation

  • SQL
  • Pandas
  • Power BI
  • Excel
More work

More projects

Supporting builds across applied ML, full-stack and scientific computing.

StreamMind AI

GenAI search and recommendation platform for a streaming catalogue.

  • Built: hybrid BM25 + semantic retrieval and a grounded RAG chatbot
  • Added: neural re-ranking and a feedback-loop experiment
  • Python
  • PyTorch
  • RAG
  • FastAPI
  • Docker
  • MLflow
Case StudyRepo available on request

CNN From Scratch

A convolutional neural network implemented in pure NumPy.

  • Built: hand-coded forward and backward passes — no frameworks
  • Covers: conv, pooling and dense layers for image classification
  • Python
  • NumPy
  • Computer Vision
Case StudyGitHub

StockScope

Full-stack stock-research platform with analytics-backed, explainable summaries.

  • Built: live search with earnings, valuation and risk breakdowns
  • Added: volatility-based forecasts and LLM-generated summaries
  • Next.js
  • TypeScript
  • Supabase
  • LLM
  • Recharts
Case StudyRepo available on request

House Price Prediction

Regression models that estimate property prices and price drivers.

  • Built: feature engineering and cross-validated model comparison
  • Result: ranked the strongest drivers of valuation
  • scikit-learn
  • pandas
  • Matplotlib
  • Python
Case StudyGitHub

Credit Risk Prediction

Classifies borrower default risk with careful imbalance handling.

  • Built: feature engineering on applicant and loan data
  • Result: interpretable risk metrics under class imbalance
  • scikit-learn
  • pandas
  • Python
Case StudyGitHub

Computational Physics (C++)

Numerical-methods toolkit in modern C++ with an N-body simulation.

  • Built: integration, interpolation and root-finding routines
  • Added: RK4 N-body simulation of the TRAPPIST-1 system
  • C++
  • Numerical Methods
  • Simulation
GitHub
Experience

Where I've worked

  1. Data Scientist · Ginseng

    Aug 2024 – Aug 2025
    • Engineered Python ML workflows using scikit-learn, XGBoost and LightGBM across 100,000+ record datasets, delivering classification, regression and clustering outputs across 6+ concurrent client workstreams.
    • Reduced analytical turnaround time by 30% by integrating LLM APIs and LangChain-assisted document processing, increasing the volume of recurring client deliverables without additional headcount.
    • Converted exploratory analysis into model evaluation reports using precision, recall, F1 and AUC, giving senior stakeholders clearer evidence for operational and strategic decisions.
    • Led a 10-person cross-functional data standardisation initiative, creating a unified reporting framework adopted across client portfolios.

  2. Finance Intern · Xceedure Business Solutions

    Jun 2024 – Aug 2024
    • Improved financial reporting accuracy by 15–20% by validating, cleansing and reconciling 30,000+ records across Excel and SQL datasets, strengthening finance data integrity before reports were issued to senior stakeholders.
    • Reduced weekly reporting preparation time from several hours to under 10 minutes by developing automated Power BI dashboards and structured Excel reporting templates for financial and operational KPI tracking.
    • Automated a recurring monthly finance reporting process using VBA macros, saving approximately 3 hours per cycle while reducing manual input errors and improving consistency across reporting outputs.
    • Strengthened spreadsheet controls using pivot tables, XLOOKUP/VLOOKUP, SUMIFS, IF statements, data validation and formula checks, improving the traceability and auditability of financial analysis.
Contact

Hiring for ML, NLP or AI roles?

I'm open to ML Engineer, AI Engineer, NLP Engineer, RAG/LLM Engineer and ML-focused Data Scientist opportunities.

Email me
EmailMuzzamil755@gmail.comLinkedInin/muzzamil-rasullyGitHubMuzzamil-Codes-dev