Yogesh Kuchimanchi | Data Science & Machine Learning

01 / About

Computerscame first.

My father brought home the first computer I ever saw. That early curiosity eventually took me from computer applications to data science and, now, an M.S. at RIT.

Research has taught me to check how labels were made, whether a comparison is fair, and where a model fails. My projects now span machine learning, public discourse, and the study of genetic evidence for type 2 diabetes.

351,501Reddit and YouTube comments analyzed

16Cases in the research corpus

3Independent windows in DriftLab

2026Paper accepted at ASONAM

NOW

Dec2026

Expected graduation, M.S. Data Science at RIT

Current study: Genetic evidence for type 2 diabetes
Recent research: Public discourse and women's safety
Paper: Accepted at ASONAM 2026
Dashboard: Open the research dashboard >

Yogesh

02 / Selected work

Selectedprojects.

Start with research and evaluation, then explore the engineering and public-data work. Each project separates what was built, what was measured, and what remains unproven.

01 //Public-discourse research

Women-Safety Public Discourse Research

An interactive research dashboard covering 351,501 Reddit and YouTube comments across 16 women-safety cases in India.

Research contribution: Python and Pandas analysis pipelines, statistical comparisons, and Qwen/LoRA model evaluation for a collaborative RIT research project.
Result: 351,501 comments across 16 cases; co-authored paper accepted at ASONAM 2026.
Evaluation discipline: The dashboard distinguishes submitted-paper results from later model audits, including limitations rather than attributing later metrics to the accepted paper.

Python
Pandas
Qwen
Statistical testing

02 //Model evaluation

DriftLab: Model Monitoring Simulator

A seeded synthetic simulator that compares an independent reference window with a current window to show when data drift does, and does not, mean model degradation.

Question: Can a monitor distinguish distribution change from model-performance degradation?
Decision: Fit the scaler and model on training data only, then keep them frozen across independent reference and current windows.
New validation: 0 false alerts in 10,000 fresh synthetic no-drift trials; two-sided 95% Wilson interval 0%–0.0384%. Fixed thresholds; no production-rate claim.

NumPy
Pandas
PSI / KS
Streamlit

03 //Product analytics · Python + SQL

Product Analytics Experimentation Lab

A reproducible synthetic experiment readout that checks whether conversion gains are sufficient to justify a rollout review.

Built: Python and SQLite analysis of users, sessions, events, and experiment assignments, with complete observation windows and explicit decision gates.
Decision: Hold. Conversion improves in the saved example, but retention uncertainty fails the declared loss-tolerance check. A positive headline metric is not enough to pass all guardrails.
Scope: 4,413 fully observed synthetic users. No measured customer benefit or real business lift is claimed.

Python
SQL / SQLite
A/B testing
Confidence intervals
Streamlit

04 //Clinical prediction

Readmission Risk Audit

An educational audit of 30-day readmission models, with patient-disjoint cohorts, probability calibration, missingness analysis, and drift checks.

Question: How useful are readmission probabilities when class imbalance, missingness, and calibration matter?
Decision: Use patient-disjoint cohorts and review PR-AUC and Brier score alongside ROC-AUC. This is an educational retrospective audit.

CatBoost
PyTorch
Calibration
Streamlit

05 //NLP

NewsSnap: News Classification and Serving

A four-topic news classifier with a FastAPI service, React dashboard, and separate transformer and keyword-demo modes.

Built: A four-topic classifier, FastAPI service, and React dashboard with separate inference modes.
Evaluation decision: Keep historical scores visible alongside training-size and text-overlap limitations. Keyword fallback is labeled rather than presented as transformer inference.

DistilBERT
FastAPI
React
CI/CD

06 //Statistical genetics · public evidence

Cardiometabolic GWAS Evidence Explorer

A public-data audit of five FinnGen type 2 diabetes regions, comparing available variant, candidate-gene, and molecular evidence while retaining source QC flags.

Question: Which candidate genes and molecular records are returned for five type 2 diabetes regions?
Decision: Preserve source QC flags, missingness, and provenance instead of presenting returned counts as evidence strength.

Python
Open Targets
Statistical genetics
GitHub Pages

07 //AI desktop tooling

Prompt Rewriter

A Windows Python tool that rewrites selected text into a structured AI prompt with a global hotkey. The case study edits one existing Notepad recording, documented as local Ollama. Cross-application reliability and inference speed have not been benchmarked.

Built: A hotkey-driven Windows workflow that rewrites selected text into a structured prompt.
Validation: A continuous Notepad demonstration preserves elapsed time. Cross-application reliability and latency remain unbenchmarked.

Python
Ollama
Desktop automation
Prompt engineering

// Repository archive

Search the project archive

Filter by project, topic, or tool

All indexed projects shown. Repository snapshot; see GitHub for the latest.

cardiometabolic-gwas-evidence-explorer >
Language: HTML. Repository source and documentation; see the README for scope and current status.
coregene-bayes-benchmark >
Language: Python. Repository source and documentation; see the README for scope and current status.
portfolio-website >
Language: TypeScript. Repository source and documentation; see the README for scope and current status.
rag-evaluation-deployment-api >
Language: Python. Repository source and documentation; see the README for scope and current status.
ai-engineering-lab >
Language: Python. Repository source and documentation; see the README for scope and current status.
NewsSnap >
Language: Python. Repository source and documentation; see the README for scope and current status.
readmission-risk-audit >
Language: Python. Repository source and documentation; see the README for scope and current status.
nyc-311-service-monitor >
Language: Python. Repository source and documentation; see the README for scope and current status.
product-analytics-experimentation-lab >
Language: Python. Repository source and documentation; see the README for scope and current status.
driftlab-model-monitoring-simulator >
Language: Python. Repository source and documentation; see the README for scope and current status.
self-healing-pipeline >
Language: Python. Repository source and documentation; see the README for scope and current status.
womens-safety-discourse-dashboard >
Language: Python. Repository source and documentation; see the README for scope and current status.
Zinga18018 >
Language: Not specified. Repository source and documentation; see the README for scope and current status.
DataGuard-AI >
Language: Python. Repository source and documentation; see the README for scope and current status.
ML-Portfolio-Projects >
Language: Python. Repository source and documentation; see the README for scope and current status.
rag-document-brain >
Language: Python. Repository source and documentation; see the README for scope and current status.
ai-code-reviewer >
Language: Python. Repository source and documentation; see the README for scope and current status.
image-captioner >
Language: Python. Repository source and documentation; see the README for scope and current status.
sentiment-engine >
Language: Python. Repository source and documentation; see the README for scope and current status.
multi-agent-orchestrator >
Language: Python. Repository source and documentation; see the README for scope and current status.
credit-card-default-prediction >
Language: Python. Repository source and documentation; see the README for scope and current status.
restaurant-analytics-llm >
Language: Python. Repository source and documentation; see the README for scope and current status.
ML-Model-Monitoring-and-Data-Drift-Detection >
Language: Python. Repository source and documentation; see the README for scope and current status.
Neural-Style-Transfer-and-Generative-Art-Lab >
Language: Python. Repository source and documentation; see the README for scope and current status.
Knowledge-Graph-Construction-and-Reasoning-Engine >
Language: Python. Repository source and documentation; see the README for scope and current status.
Time-Series-Forecasting-with-Transformer-Architectures >
Language: Python. Repository source and documentation; see the README for scope and current status.
Multimodal-Anomaly-Detection-with-Synthetic-Augmentation >
Language: Python. Repository source and documentation; see the README for scope and current status.
Ipynb_Project_Snippets >
Language: Jupyter Notebook. Repository source and documentation; see the README for scope and current status.
SymptomAid-AI >
Language: Python. Repository source and documentation; see the README for scope and current status.
stock-sentiment-dashboard >
Language: Python. Repository source and documentation; see the README for scope and current status.
MetalVision-AI >
Language: Python. Repository source and documentation; see the README for scope and current status.
Hecs-Refactoring >
Language: Python. Repository source and documentation; see the README for scope and current status.
Job-Hustles >
Language: Not specified. Repository source and documentation; see the README for scope and current status.
imdb-sentiment-analysis-deep-learning >
Language: Not specified. Repository source and documentation; see the README for scope and current status.
Wispr-Clone >
Language: Python. Repository source and documentation; see the README for scope and current status.
NeuroCanvas >
Language: Not specified. Repository source and documentation; see the README for scope and current status.
Project-Aegis >
Language: Python. Repository source and documentation; see the README for scope and current status.
NeuroCanvas-AI-Art-Generator >
Language: Python. Repository source and documentation; see the README for scope and current status.
time-chronicles-ai >
Language: JavaScript. Repository source and documentation; see the README for scope and current status.
ai-slide-deck-generator >
Language: HTML. Repository source and documentation; see the README for scope and current status.
echo-muse-therapeutic-storytelling >
Language: Python. Repository source and documentation; see the README for scope and current status.
sonic-symbiosis >
Language: HTML. Repository source and documentation; see the README for scope and current status.
mycelial-memories >
Language: JavaScript. Repository source and documentation; see the README for scope and current status.
Python-Script >
Language: Python. Repository source and documentation; see the README for scope and current status.
ollama-image-recognition-tool >
Language: HTML. Repository source and documentation; see the README for scope and current status.
ml-iris-data-analysis >
Language: Python. Repository source and documentation; see the README for scope and current status.
ml-digits-exploration >
Language: Python. Repository source and documentation; see the README for scope and current status.
customer-churn-prediction >
Language: Python. Repository source and documentation; see the README for scope and current status.

More work on GitHub.

Find the source, setup instructions, and evaluation notes for more of my projects.

Synthetic Bayesian benchmark: coregene-bayes-benchmark
RAG evaluation API: rag-evaluation-deployment-api
AI engineering lab: ai-engineering-lab
Product analytics and A/B testing: product-analytics-experimentation-lab
Service monitoring: nyc-311-service-monitor

Browse GitHub repositories

Interactive project detail

Explore thesource evidence.

A closer look at the five-region genetics project. Select a region to inspect returned records and their limits.

Genetics project / Five selected regions

Evidence atlas

FinnGen R12 T2D · Open Targets 26.09

Select a type 2 diabetes region 01 / 05TCF7L2 02 / 05CCND2 03 / 05FTO 04 / 05HNF4A 05 / 05CDKAL1

Lead variant location

chr10:112,998,590

Top returned L2G candidate: TCF7L2 · not a causal-gene call

Fine-mapping → candidate ranking

Variant rows: 1
L2G candidate rows: 1

Parallel molecular-QTL record

Colocalisation rows: not returned
Candidate-matched rows: not returned

Lead variant location

chr12:4,275,678

Top returned L2G candidate: CCND2 · not a causal-gene call

Fine-mapping → candidate ranking

Variant rows: 1
L2G candidate rows: 2

Parallel molecular-QTL record

Colocalisation rows: 38
Candidate-matched rows: 11

Lead variant location

chr16:53,793,567

Top returned L2G candidate: FTO · not a causal-gene call

Fine-mapping → candidate ranking

Variant rows: 56
L2G candidate rows: 3

Parallel molecular-QTL record

Colocalisation rows: 2
Candidate-matched rows: not returned

Lead variant location

chr20:44,354,839

Top returned L2G candidate: HNF4A · not a causal-gene call

Fine-mapping → candidate ranking

Variant rows: 1
L2G candidate rows: 4

Parallel molecular-QTL record

Colocalisation rows: not returned
Candidate-matched rows: not returned

Lead variant location

chr6:20,703,721

Top returned L2G candidate: CDKAL1 · not a causal-gene call

Fine-mapping → candidate ranking

Variant rows: 6
L2G candidate rows: 1

Parallel molecular-QTL record

Colocalisation rows: 4
Candidate-matched rows: 2

Counts show returned-record availability, not evidence strength. “Not returned” is missingness, not a negative biological finding. L2G means locus-to-gene; its score may already include colocalisation, so the molecular track is not independent validation.

All five source regions carry an unspecified quality-control flag.Open full explorer >

03 / Experience

Researchand work.

Aug 2025 - Present

Graduate Research Project, Data Science

Rochester Institute of Technology

Built Python and Pandas pipelines to analyze 351,501 Reddit and YouTube comments across 16 women-safety cases in India.
Applied Mann-Whitney U, chi-square, and G-tests; fine-tuned Qwen3.5-9B with LoRA and reviewed a fixed 3,000-row validation sample.
Co-authored a paper accepted at ASONAM 2026 and documented model limitations through high-confidence error review.

Aug 2025 - Present

GCCIS Technical Assistant

Rochester Institute of Technology

Support 200+ Windows and Linux lab systems, troubleshoot access and software issues, and maintain Excel/VBA reports.

Dec 2024 - Apr 2025

Research Assistant

Rochester Institute of Technology

Refactored Pandas and NumPy scripts and validated PostgreSQL and Excel data for accurate weekly research reporting.

Education

Aug 2024 - Dec 2026

M.S. in Data Science

Rochester Institute of Technology · Rochester, NY

Expected December 2026. Machine Learning, Deep Learning, Cloud Computing, Big Data Analytics.

Aug 2023 - Jul 2024

Postgraduate Program in Data Science

Vellore Institute of Technology · India

Nov 2020 - Dec 2023

B.C.A. in Computer Applications

New Shores International College · India

04 / Skills

ToolsI use.

Languages, modeling libraries, databases, and tools used in my projects.

Languages

Python
SQL
R
Java

Machine Learning & NLP

See NLP research >

PyTorch
scikit-learn
CatBoost
XGBoost
Transformers
LoRA
DistilBERT
FT-Transformer

Statistics & Evaluation

See evaluation decisions >

A/B testing
Mann-Whitney U
Chi-square
G-test
ROC-AUC
PR-AUC
Calibration
PSI / KS / JS

Data & Databases

See data-quality audit >

Pandas
NumPy
SciPy
Spark
PostgreSQL
SQLite
Window functions

ML Engineering

See serving architecture >

FastAPI
Streamlit
React
Docker
GitHub Actions
Automated testing
Git
Excel / VBA

05 / Contact

Get intouch.

I'm seeking early-career data science and ML engineering roles. Expected graduation: December 2026. Email me about a role or research collaboration; I'm happy to share my résumé and discuss the decisions behind these projects.

// EMAILkuchimanchiyogesh@gmail.com

// GITHUBgithub.com/Zinga18018 >

// LINKEDINlinkedin.com/in/yogesh-kuchimanchi >

// SUMMARYRead / print career summary >

// RÉSUMÉRequest my résumé >

// BASED INRochester, New York

Email me