Data Scientist & AI Engineer

Bijaya Kumar Pariyar

Building production-grade AI systems | RAG pipelines | LLM integrations | Data science at scale

View Projects Contact Me

20+

ML Projects

1M+

Rows Processed

50+

GitHub Repos

About Me

Self-taught Data Scientist with hands-on experience building end-to-end ML solutions across classification, regression, clustering, anomaly detection, and time series forecasting.

Currently working as a Data Scientist Intern at CR Equity AI, Inc., where I design and implement production-grade RAG pipelines, integrate LLM APIs, and build FastAPI services for document intelligence. My work spans from classical ML to cutting-edge AI systems, with a strong focus on security, scalability, and real-world deployment.

I've completed over 20 structured ML projects, handling large-scale datasets with 1M+ rows for fraud detection and 1.7M+ rows for forecasting. My approach combines rigorous evaluation, model explainability through SHAP, and reproducible research workflows.

Education: BCA, Tribhuvan University
Location: Kathmandu, Nepal
Role: Data Scientist Intern
Company: CR Equity AI, Inc.
Quick Links
GitHub LinkedIn Kaggle Email

Experience

Data Scientist Intern

CR Equity AI, Inc.

Remote (Tallahassee, Florida, United States)

Dec 2025 - Present

Contributing to LLM-based RAG systems and backend AI services for document intelligence. Collaborating remotely with a US-based engineering team using Agile development practices.

Key Responsibilities & Achievements:
  • Designed and implemented end-to-end RAG pipelines including document chunking, embedding generation, vector retrieval, reranking, and prompt construction
  • Integrated and managed LLM APIs with retry logic, timeout handling, prompt/token constraints, and response validation
  • Built production-grade FastAPI services for document processing, querying, and AI inference
  • Worked with vector databases (FAISS, Qdrant, PostgreSQL pgvector) for scalable semantic search
  • Implemented asynchronous processing pipelines using Redis-backed job tracking and webhook-based callbacks
  • Applied security best practices including JWT authentication, SSRF protection, rate limiting, CORS, and trusted host validation
  • Containerized backend services using Docker and implemented structured logging, metrics collection, and observability
  • Conducting research on SOC practices and security hardening for production AI and backend systems
Python FastAPI RAG LLMs FAISS Qdrant Docker Redis PostgreSQL JWT

Featured Projects

A selection of my recent work in ML, AI, and data science

Featured

Personal Knowledge Base RAG API

FastAPI • FAISS • Qdrant • Groq • LLMs

Modular RAG API built from scratch with FastAPI for document upload and natural language querying. Features switchable vector backends (FAISS/Qdrant), multiple chunking strategies, cross-encoder reranking, and Groq-powered generation with Llama-3.3-70B.

  • FastAPI endpoints for TXT/PDF ingestion and semantic querying
  • Configurable chunking strategies (fixed-size, sentence, recursive)
  • Thread-safe writes with metadata filtering
  • Low-latency responses with sentence-transformers embeddings
View on GitHub
RAG Precision Enhancement

FAISS • Cross-Encoder • Reranking

Comparative study demonstrating why retrieval alone isn't enough. Built two-stage retrieval + reranking pipeline with noise injection to stress-test semantic search performance.

  • Noise injection with 20+ hard-negative documents
  • Cross-encoder reranking (BAAI/bge-reranker-base)
  • Evaluated via Hit Rate@k and MRR metrics
GitHub
ChurnShield

Flask • Scikit-learn • SHAP • SQLite

Flask web app for real-time churn prediction with authentication, admin dashboard, and CSV export. Integrated Random Forest pipeline with SHAP explainability.

  • Real-time prediction with Random Forest
  • SHAP-based feature importance
  • Automated retention strategy generator
GitHub
IEEE Fraud Detection

XGBoost • SHAP • Imbalanced Learning

Transaction classification on 1M+ Kaggle IEEE-CIS records using XGBoost. Achieved ROC-AUC 0.95 and F1 0.66 on imbalanced fraud class.

  • Processed 1M+ transaction records
  • Feature reduction via Spearman correlation
  • Hyperparameter tuning with Optuna
GitHub
Time Series Forecasting

LightGBM • Prophet • Feature Engineering

SKU-level supermarket price forecasting on 1.7M+ rows using LightGBM and Prophet with lag, rolling, and calendar features.

  • Processed 1.7M+ retail records
  • TimeSeriesSplit cross-validation
  • Evaluated via RMSE, MAPE, SMAPE
GitHub
Retail Price Optimization

Random Forest • Optuna • SHAP

Competition-aware ML workflow integrating historical sales and competitor pricing. Achieved R² ≈ 0.92 with Random Forest and SHAP explainability.

  • Full ML pipeline with EDA
  • Hyperparameter tuning with Optuna
  • Profit and revenue optimization
GitHub
Anomaly Detection

Isolation Forest • LOF • KMeans • DBSCAN

Multi-method anomaly detection pipeline combining model-based, cluster-based, and statistical approaches on fraud and retail datasets.

  • Multiple detection algorithms compared
  • Evaluation via ROC-AUC and precision/recall
  • PCA visualization of anomalies
GitHub
More Projects
Regression Portfolio Customer Segmentation Text Embeddings Explorer Semantic Search (FAISS) Qdrant Vector DB Text Chunking Strategies

Technical Skills

Programming
Python SQL Java JavaScript PHP
Machine Learning
Classification Regression Clustering Anomaly Detection Time Series Feature Engineering Hyperparameter Tuning SHAP
AI & NLP
LLMs RAG Systems Vector Databases Embeddings Semantic Search Reranking
Frameworks & Libraries
Scikit-learn Pandas NumPy LightGBM XGBoost Prophet sentence-transformers
Databases & Vector Stores
FAISS Qdrant PostgreSQL (pgvector) Redis SQLite MySQL
Deployment & DevOps
FastAPI Flask Docker Git/GitHub JWT Auth Rate Limiting

Research

SHAP-Based Feature Selection and Iterative Hyperparameter Tuning for Customer Churn Prediction in Telecommunication Datasets

Preprint • 2024

Explores interpretability and optimization techniques for churn modeling, demonstrating SHAP values for transparent feature importance and iterative tuning for enhanced model performance on real-world telecom data.

SHAP Hyperparameter Tuning Churn Prediction Telecom

Get In Touch

I'm always interested in discussing new opportunities, collaborations, or innovative projects

Email

bijaybeezoe@gmail.com

Phone

+977-9767645335

Location

Kathmandu, Nepal