Supply Chain · Machine Learning · Agentic AI
FlowForge
- Role
- Personal technical project
- Date
- 2026
Kedroscikit-learnLangChainOpenAI
UC Berkeley · EPFL
Thibault Verny
Statistical Learning, Agentic AI & Real-World Systems
M.A. student in Statistics & Data Science at UC Berkeley, following a B.Sc. in Mechanical Engineering at EPFL.
My work combines statistical learning, machine learning, and agentic AI to build reliable decision systems for real-world problems.
Available for full-time roles from May 2027. Open to conversations before then.
Berkeley, California
Areas of focus
The technical areas connecting my projects and experience.
My work spans statistical learning, uncertainty, intelligent systems, and technical decision tools, with a focus on methods that remain useful and reliable on imperfect real-world data.
Statistical Learning & Uncertainty
Probabilistic modeling, robust inference, calibration, validation, and uncertainty quantification.
Agentic AI & Decision Systems
Tool-using LLM systems, deterministic analytics, agent orchestration, evaluation, and decision-support pipelines.
Sensing, Localization & Control
Signal processing, sensor fusion, localization, simulation-to-hardware validation, and feedback control.
Featured project
FlowForge: Agentic Supply Chain Decision Intelligence
Supply Chain · Machine Learning · Agentic AIPersonal technical project2026
End-to-end supply-chain decision system combining reproducible Kedro ML pipelines, leakage-safe late-delivery modeling, deterministic business analytics, and LangChain-based decision support.
Project scope
- Kedro pipeline on the DataCo order dataset
- Leakage-safe features at order-creation time
- Chronological 70 / 15 / 15 train / validation / test split
- Logistic Regression late-delivery probability model
- Revenue-weighted exposure ranking
- Deterministic analytics tools orchestrated by LangChain + OpenAI
- Streamlit Control Tower with Flow View and operational brief
My contributions
- Built the Kedro pipeline from 180,519 order-item rows to 65,752 order-level records, with Parquet intermediates and pandas.
- Implemented leakage-safe order-creation features and a chronological train / validation / test split (final test slice: 9,863 orders).
- Fit a Logistic Regression baseline and reported final chronological test metrics (ROC-AUC 0.7490, Brier Score 0.1967).
- Shipped a Streamlit Control Tower: KPI cards, geographic store-coordinate map, Flow View Sankey, shipping-mode and regional tables, and a LangChain operational brief over deterministic tools.
Pipeline
- 01Raw orders
- 02Kedro pipeline
- 03P(late delivery)
- 04Revenue exposure
- 05Deterministic tools
- 06Operational brief
Projects
Selected projects
Project 1 of 6: FlowForge
Background
Education and experience
Two separate tracks: academic education, and professional, research, and leadership experience.
Education
University of California, Berkeley
Aug 2026 – May 2027 (expected)M.A. Statistics & Data Science
Berkeley, CaliforniaCurrent graduate student
École Polytechnique Fédérale de Lausanne (EPFL)
Sep 2023 – Jul 2026B.Sc. Mechanical Engineering
Lausanne, Switzerland
Activities & leadership
Vice-President, Junior Enterprise EPFLWindsurfing InstructorHead of Sailing, SPI LausanneFirst-Year Student Coach
Selected quantitative coursework
- Calculus III5.75 / 6.0
- Calculus IV5.50 / 6.0
- Probability and Statistics5.50 / 6.0
- Programming for Engineers5.50 / 6.0
- Dynamical Systems5.50 / 6.0
- Optimization and Operations Research5.00 / 6.0
- Control Systems + Lab5.00 / 6.0
EPFL grading scale: 6.0 is the highest possible grade; 4.0 is the passing grade.
Experience
Wakam
Jun 2026 – Aug 2026AI Engineer Intern
IndustryBuilt the governance and analytics layer for an internal LLM agent platform: ingestion workflows into PostgreSQL, monthly snapshots, and agent lifecycle, ownership and freshness signals feeding review recommendations rather than automated deletions. Also delivered credit-consumption analytics by period, user and agent, and prototyped a semantic similarity pipeline for agent configurations using deterministic fingerprinting and vector retrieval.
EPFL Automatic Control Laboratory
Feb 2026 – Jun 2026Bachelor's Thesis Research Project
Research ProjectDeveloped and experimentally validated a UWB-based localization and control system for robotic tracking, combining physical sensing, empirical noise modeling, ROS 2 simulation, and hardware integration.
EPFL
Sep 2025 – Dec 2025Teaching Assistant — Calculus III & Programming for Engineers
TeachingLed problem-solving sessions and supported undergraduate students in advanced calculus, scientific computing, Python, C, and MATLAB.
DPULSE.AI
Sep 2025 – Jan 2026Co-founder & Founding Engineer
EntrepreneurshipDeveloped probabilistic NLP pipelines for organizational narrative analysis, with a focus on uncertainty and responsible interpretation.
Junior Enterprise EPFL
Sep 2024 – May 2026Vice-President
LeadershipScoped and delivered a CHF 150k+ portfolio of ML/AI and software mandates directly with external clients, from first scoping conversation through delivery, in a year when Junior Enterprise EPFL reached a record CHF 425k and was named Best Junior Enterprise in Switzerland 2026.
Organizational recognition: Best Junior Enterprise in Switzerland 2026 — Junior Enterprises Switzerland
Capabilities
Technical breadth
Tools and methods I work with across statistics, software, and real-world systems.
AI & Agentic Systems
- OpenAI API
- LangChain
- Tool calling
- Structured outputs
- MCP
- Langfuse
- Cursor
- Claude CLI
Data Science & ML Engineering
- Python
- SQL / PostgreSQL
- Kedro
- pandas / Polars
- NumPy / SciPy
- scikit-learn
- Statistical modeling
- Supervised learning
- Unsupervised learning
- Forecasting
- Clustering
- Optimization
- Simulation
- Model evaluation
Software & Product
- FastAPI
- TypeScript
- REST APIs
- ETL pipelines
- Docker
- Git
- CI/CD
- MLOps / LLMOps
- Next.js
- React
- React Native
- Streamlit
- Supabase
Quantitative Systems
- Signal processing
- Control
- ROS 2 / Gazebo
- Simulation
- Experimental validation
Contact
Let’s build reliable systems for real-world data.
Open to technical conversations and opportunities involving data science, applied machine learning, agentic AI, quantitative modeling, and real-world decision systems.