Skip to contentSadiksha Chitrakar

Sadiksha Chitrakar

Data Scientist · AI / ML Engineer

I build AI that reasons out loud.

Retrieval and agentic assistants, applied ML, and the pipelines behind them — built end to end, with an engineer’s rigor and a researcher’s honesty. I turn models into systems you can open, run, and trust.

campusguide · agentic rag
0.00ssimulated
Question
Route
Retrieve
Generate
Verify
Answer
run.log
Watching the pipeline route, retrieve, and verify…

Open any project below and run the workflow live

About

A data scientist who builds the whole system.

Sadiksha Chitrakar
Location
Miami, Florida, USA
Studying
MS Data Science · UNH ’26

I’m a data scientist and AI/ML engineer, and I build across the whole arc of a system — data, models, reasoning, and the interface someone actually uses. I’ve built agentic RAG assistants that route a question, pull the right evidence, and check their own answers; applied-ML and computer-vision models trained and evaluated end to end; and serverless data pipelines on AWS that turn raw data into forecasts and dashboards.

Alongside that I’ve worked on healthcare ML research, and I came to data science from two years as a software engineer — the production rigor I bring to everything now. That mix of research and engineering is why my projects tend to become systems you can actually open and use, not notebooks that stay on a shelf.

Education
  • University of New Haven
    MS, Data ScienceUniversity of New HavenAug 2024 – May 2026 · GPA 3.97
  • Coventry University
    BSc (Hons), Computing / Computer ScienceCoventry UniversityOct 2019 – Nov 2022 · GPA 3.94
Certifications

Power BI — Data Modeling & Visualization · Apache Spark — Databricks · MLOps Essentials: Model Development & Integration

Experience
  1. University of New Haven

    Graduate Research Assistant · University of New Haven

    Aug 2025 – Dec 2025

    West Haven, CT

    MAE 3.5 yr · R² 0.95 on 75k+ individuals

    Applied data science research on large-scale public health data (NHANES).

    • Built preprocessing and cohort-construction pipelines over NHANES data (75,000+ individuals) across multiple survey cycles — MICE imputation, standardization, and feature engineering.
    • Developed and evaluated models to estimate biological age from clinical and biomarker data with 5-fold cross-validation; XGBoost reached MAE 3.5 years (R² 0.95), outperforming Random Forest by ~18% lower MAE.
    • Used feature-importance analysis to surface the biomarkers driving the model.
  2. Xuno

    Software Engineer · Xuno

    Jan 2024 – Jan 2025

    Ellensburg, WA

    ~30% fewer post-deployment bugs

    Cross-border money-transfer product, contributing to architecture and reliability.

    • Contributed to system architecture and design decisions — weighing trade-offs in data flow, performance, and scalability to improve reliability and long-term maintainability.
    • Built unit, widget, and integration testing strategies using logs and runtime metrics, catching defects early and reducing post-deployment bugs by ~30%.
  3. Code Himalaya

    Flutter Developer · Code Himalaya

    Nov 2022 – Jan 2024

    Lalitpur, Nepal

    ~70% faster releases · 30+ apps shipped

    Mobile product development and release automation.

    • Developed and maintained Flutter UI components, integrating REST APIs and structured data models, validating data mapping and handling edge cases to improve stability.
    • Automated build and deployment with Fastlane CI/CD — cutting release-cycle time by ~70% and publishing 30+ apps to the Play Store and App Store.
Work

The systems I build — and proof they work.

Open any project to see the architecture and results — and run the interactive ones live.

Retrieval & agentic assistants

Systems that route a question, retrieve and rank the right evidence, grade their own grounding, cite sources, and escalate when the evidence falls short.

Applied machine learning

End-to-end modeling — preprocessing, feature engineering, model comparison, and honest cross-validated evaluation, from tabular healthcare data to computer vision.

Data & cloud pipelines

Serverless ingestion, ETL, and analytics on AWS that turn raw data into forecasts and dashboards.

Selected projects
RAG System

CampusGuide

Agentic RAG assistant for university information

An agentic retrieval assistant that answers campus questions by routing each one, searching documents and structured data, grading its own evidence, verifying the answer, and escalating when the sources aren’t enough.

7/8Route accuracyon the repo’s eval set
  • Python
  • LangGraph
  • LangChain
  • ChromaDB
  • MPNet embeddings
  • +4
Agentic workflow
  1. Question received
  2. Route the question
    route_question
  3. Call the right tool
    search_campus_docs · lookup_*
  4. Grade the evidence
  5. Rewrite & retry
    conditional
  6. Generate grounded answer
Machine Learning

Biological Age from Biomarkers

Healthcare ML research on NHANES data

A research-oriented modeling study that estimates biological age from clinical and biomarker data across multiple NHANES survey cycles, comparing models with honest cross-validated evaluation.

3.5 yrMAEXGBoost · 5-fold CV
  • Python
  • XGBoost
  • Random Forest
  • scikit-learn
  • Pandas
  • +3
Model pipeline
  1. NHANES survey cycles
  2. Cohort construction
  3. Preprocess
    MICE · standardization
  4. Feature engineering
  5. Train & compare models
    XGBoost · RandomForest
  6. Cross-validated evaluation
    5-fold CV
Machine Learning

EquipSense

Predictive maintenance from machine sound

An audio-based predictive-maintenance assistant: it listens to a pump recording, flags abnormal behavior, retrieves similar past maintenance reports, and drafts a technician-ready recommendation.

74-dAudio featuresMFCC · chroma · spectral · ZCR · RMS
  • Python
  • Streamlit
  • Librosa
  • scikit-learn
  • Random Forest
  • +3
Model pipeline
  1. Recording loaded
  2. Extract audio features
    Librosa
  3. Classify condition
    Random Forest
  4. Anomaly score
    IsolationForest
  5. Retrieve similar reports
    TF-IDF cosine
  6. Draft recommendation
    rule-based / Gemini
Data Science

Sales Forecasting Pipeline

Serverless data engineering on AWS

A serverless AWS pipeline that ingests multi-store retail data, runs ETL, catalogs it for SQL analytics, trains a demand-forecasting model, and serves forecasts to a Power BI dashboard.

4Lambda functionspreprocess · load · catalog · serve
  • AWS S3
  • AWS Lambda
  • Amazon RDS
  • AWS Glue
  • Amazon Athena
  • +3
Data pipeline
  1. Upload to S3
    Amazon S3
  2. Preprocess
    AWS Lambda
  3. Load to RDS
    Amazon RDS
  4. Crawl & catalog
    AWS Glue
  5. Query
    Amazon Athena
  6. Forecast demand
    scikit-learn
Computer Vision

Building-Component Keypoint Detection

Computer vision with Keypoint R-CNN

A Keypoint R-CNN model that detects doors, windows, and roofs in images and regresses their semantic keypoints — trained with transfer learning on a custom, augmented COCO dataset.

0.89Detection mAPheld-out test set
  • PyTorch
  • Torchvision
  • Keypoint R-CNN
  • ResNet-50 + FPN
  • Albumentations
  • +2
Vision pipeline
  1. COCO dataset
  2. Augment
    Albumentations
  3. Keypoint R-CNN
    PyTorch
  4. Transfer-learn
  5. Evaluate
    OKS · mAP
  6. Predict & visualize
Automation

Tango

Constraint-validation game + difficulty ML

A constraint-based logic puzzle with a live constraint-validation engine, paired with a small ML study that predicts a puzzle’s difficulty from its parameters.

215Games collectedplayed among friends
  • Python
  • Pygame
  • NumPy
  • scikit-learn
  • K-Means
  • +1
Automation flow
  1. Constraints
  2. Validate state
  3. Predict difficulty
    K-Means · Random Forest
  4. Easy / Medium / Hard
Toolkit

The stack I reach for.

Tools I’ve used in projects and research.

Languages
  • Python
  • SQL
  • R
  • JavaScript
  • Dart
AI / ML
  • RAG
  • LangChain
  • LangGraph
  • LLMs
  • Agentic AI
  • MCP
  • NLP
  • Computer Vision
  • Prompt Engineering
  • scikit-learn
  • PyTorch
  • TensorFlow
  • XGBoost
  • HuggingFace Transformers
  • OpenCV
  • Feature Engineering
  • Model Evaluation
Data & Cloud
  • Pandas
  • NumPy
  • ETL
  • Data Modeling
  • Apache Spark
  • AWS (S3 · Lambda · RDS · Glue · Athena · SageMaker · API Gateway)
  • PostgreSQL
  • MySQL
  • ChromaDB
  • Firebase
  • MongoDB
Tools
  • FastAPI
  • Streamlit
  • Docker
  • REST APIs
  • CI/CD
  • Power BI
  • Tableau
  • Excel
  • Git
  • GitHub
  • Jira
Available for roles & collaborations

Let’s build something useful.

Recruiters, hiring managers, and collaborators — I’d love to hear what you’re working on. The fastest way to reach me is email.

[email protected]LinkedInGitHub