Tchifou M. Dieffi — Data Scientist & ML Engineer

01 — About

What I care about

My research centers on understanding what ML models actually do rather than taking their outputs at face value — from SHAP and KernelSHAP interpretability work to studying what PCA-based debiasing techniques really remove from word embeddings.

Professionally, I've worked as a data analyst on enterprise data warehousing, ELT pipelines, and Medicare claims data, and I build ML applications end to end — from model training through deployment.

02 — Work

Featured projects

digitix↗

Full-stack CNN application for handwritten digit recognition, with a live image-processing pipeline and a Streamlit interface.

PythonCNNStreamlit

i2g↗

Open-source Python library, published on PyPI, converting grayscale images into graph structures for GNN modeling.

PythonPyPIGNN

shap_stability_framework↗

Benchmarking framework diagnosing and resolving algorithmic instability in KernelSHAP explanations.

PythonXAI

SyndromeX↗

Clinical decision support web application for real-time syndrome identification in epidemiological surveillance.

HealthcareWeb App

aws-lake-house↗

Cloud data lakehouse pipeline curating ML training data using AWS Glue, Athena, and Spark.

AWSSparkETL

TellMeMore↗

Full-stack app (Python backend, Vue.js frontend) for structured benchmarking and evaluation of LLM outputs across OpenAI, Gemini, and Groq LLaMA models. Live at deepoctopus.com.

PythonVue.jsLLMs

03 — Skills

Toolbox

AI / ML

TensorFlowPyTorchscikit-learnSHAPLIMEPredictive Modeling

LLM & NLP

OpenAIGeminiClaude APIPrompt Engineering

Languages

PythonSQLPL/SQLJavaScriptC++

Data Engineering

ETL PipelinesData ValidationQuality ControlSnowflake

Data & Visualization

Power BITableauMatplotlibSeabornExcel

Tools & Cloud

GitDockerLinuxNode.jsVue.jsAWSGCP

04 — Education

Education & publication

M.S. in Data Science

Hood College — Expected Fall 2026

Thesis: Explainable AI (SHAP) + LLM integration for predictive model transparency.

Postgraduate Certificate, Data Science

University of Texas at Austin — Oct 2022

IBM Data Analyst Professional Certificate

Oct 2021

B.S. in Information Technology

African Institute of Computer Science

PUBLICATION

What Does Debiasing Really Remove? A Geometric Study of PCA-Based Gender Debiasing in Word Embeddings

Kresin, A., Tchifou M. Dieffi, & Caspi, T. (2026) · arXiv:2606.07964 · Hood College & Ben-Gurion University ↗

05 — Experience

Where I've worked

Graduate Researcher

Hood College · Frederick, MD

2024 — Present

Built SyndromeX, a real-time clinical decision support system for syndrome detection from structured patient data, designed for operational transparency and executive reporting.
Developed a web-based explainable AI system (SHAP + LLMs) translating predictive model outputs into interpretable summaries for non-technical stakeholders — core of MS thesis.
Co-authored a peer-reviewed arXiv paper on geometric analysis of gender debiasing in NLP word embeddings, a cross-institutional collaboration with Ben-Gurion University.

Data Engineer / Data Analyst

DIMCORP LLC · Frederick, MD

2020 — Present

Designed and managed enterprise databases and automated ETL pipelines integrating mobile, web, and external APIs for real-time reporting across global health programs in multiple countries.
Built recurring decision-support dashboards and reports for operational and program leadership, keeping data definitions and methodologies consistently documented.
Implemented structured data validation, quality control, and system testing frameworks that improved data reliability for downstream analysis.
Led development of mobile and web data collection platforms across multi-country deployments, providing technical mentorship to local teams.