AI/ML Engineer • Probabilistic Modeling • Deep Learning • Optimization

Hi, I'm Ajeenckya Mahadik

I design and train machine learning systems with a focus on probabilistic modeling and decision-making — from calibrated prediction engines to stochastic optimization and computer vision.

Graduating May 2026 from UW–Madison (M.S. Industrial & Systems Engineering), currently building a probabilistic outcome-simulation system.

View Current Project Download Resume

GitHub

About

AI/ML engineering with a solid optimization and probabilistic modeling backbone.

I'm a Master's student in Industrial & Systems Engineering at the University of Wisconsin–Madison, specializing in machine learning, stochastic modeling, and optimization.

I like taking messy, high-dimensional problems and turning them into end-to-end ML systems: calibrated probability models, simulation-ready outputs, and pipelines that support fast iteration.

My current project focuses on a Probabilistic AI System for Outcome Simulation, designed to generate stable probability distributions and support robust scenario analysis.

What I enjoy working on

Probabilistic ML
Prediction & Simulation
Stochastic modeling
Optimization
Computer Vision
LLM-based feature engineering

Skills & Tools

Core areas I use to design, train, and analyze models.

Representation Learning & Text Features

Extracting meaningful signals from news, text, and metadata for downstream ML systems.

Embeddings LLM features Text preprocessing

Machine Learning & Deep Learning

Building prediction systems for structured data, time series, and vision tasks.

PyTorch scikit-learn CNNs Transformers

Probabilistic Modeling

Calibration, uncertainty estimation, Monte-Carlo evaluation, and scenario forecasting.

Uncertainty estimation Calibration Prediction intervals

Optimization & Systems

Deterministic and stochastic optimization with real-world constraints.

Gurobi Routing MIP / MILP

Projects

My ongoing and upcoming machine learning work.

Probabilistic AI System for Outcome Simulation

A prediction engine that generates calibrated probability distributions instead of single hard outputs. It analyzes historical patterns, estimates likely outcomes, and measures how confident the system is in each scenario.

Designed for forecasting and decision-making under uncertainty: it shows how outcomes shift as conditions change and how stable predictions remain across repeated simulations.

Probability modeling Simulation Forecasting Risk analysis

GitHub (code & experiments)

Upcoming Projects

Concrete next phases extending my probabilistic prediction system.

News-Driven Behavioral Signal Modeling

A microservice that turns real-time news into structured behavioral vectors using lightweight LLMs. These features improve probability calibration and directional accuracy.

Early prototypes improve calibration by ~12% and strengthen scenario reasoning.

LLM embeddings Feature engineering Microservice

Weather-Impact Prediction Layer (MDN)

A Mixture Density Network designed to model environmental influences—temperature, humidity, rainfall, elevation—on outcome distributions.

Reduces scenario variance by ~17% and helps quantify environmental uncertainty.

Mixture Density Networks Environmental modeling

CTMC Dynamics for Time-Evolving Simulation

A Continuous-Time Markov Chain (CTMC) layer to track dynamic shifts in conditions throughout simulated scenarios.

Enables richer multi-step forecasting and evolving probability curves.

CTMC Time dynamics

Future Ideas

High-level concepts to explore next.

Multi-Agent Simulation

Concept for agents acting under uncertainty to study cooperation, stability, and complex decision behavior.

LLM Behavior & Safety Modeling

A lightweight predictor that estimates the risk of an LLM response before generation completes—enabling proactive filters and safer applications.

Other Projects

Additional work in optimization, vision, and forecasting.

Adpative Two Stage Stochastic Routing

A two-stage stochastic routing model with recourse to minimize expected cost under uncertain travel times.

Routing Stochastic programming

GitHub

Facial Expression Detection

A ResNet50 + Vision Transformer hybrid for low-resolution emotion recognition.

Computer vision ResNet50 ViT

GitHub

Mulitmodal Route Optimization

A constrained multimodal routing framework that computes cost–time optimal travel paths across bus, train, and flight networks.

Pyomo Travel Salesman Problem

GitHub

Experience

Where I’ve applied engineering and analytical thinking so far.

Graduate Research Assistant — UW–Madison

Working on probabilistic modeling, simulation stability, and optimization-based ML systems supporting research workflows.

FIAT India Automobiles — Graduate Apprentice Trainee

Improved powertrain launch operations through cycle-time analytics and constraint-driven line balancing, increasing assembly efficiency by 9%. Built digital procedures and engine-assembly workflows that enabled traceable metrics and automation-ready process data.

Powertrac Tractors — Maintenance Engineer

Applied reliability analysis to reduce downtime patterns and improve technician productivity by 12%. Designed engineered jigs, gauges, and optimized layouts to generate cleaner operational signals and more consistent maintenance data for future predictive systems.

Resume

One-page snapshot of my AI/ML background.

Download my resume for a concise view of my projects, experience, and technical skills tailored for AI/ML Engineer roles.

Open Resume (PDF)

Contact

Best ways to reach me.

GitHub

Email me about an AI/ML role