AI & Data Analyst · Boston, MA

From trade floors
to training sets.

same curiosity.

An AI & Data Analyst with an MS in Business Analytics from UMass Amherst. Before grad school I worked in international trade: delegations, market research, and a lot of data nobody quite trusted. So I learned to fix it, then to model it. Now I build the model, test it, and make the answer obvious.

Madhura in her graduation gown and cap, sitting among flowers
MS Business Analytics · UMass Amherst · 2026 · GPA 3.9

Analyse.

SQL, Python and R on messy, real-world data: cleaned, reconciled and tested before anyone is asked to trust it.

Model.

Predictive models tuned to the cost of being wrong, not just to the accuracy score.

Translate.

Dashboards and briefings that get the decision made in the meeting, not three emails later.

06 projects

Selected
work

Six industries, six decisions. Open any project for the problem, what I did, and what it changed.

FinanceFraud detection
HospitalityPrice modelling
HealthcareRisk prediction
TravelSegmentation
LogisticsNetwork design
RetailDashboards
  • operating pointAUC0.978false positive rate →true positive rate →577 : 1 IMBALANCE
    FinanceAI & LLM

    Fraud triage with LLM risk narratives

    0.978AUC
    84%Recall
    284,807Transactions
    577:1Imbalance
    Problem
    284,807 card transactions with 577 legitimate for every fraudulent one, where a model that flags nothing still scores 99.8% accuracy.
    Approach
    Two-stage detection with weighted loss functions so the model is punished for missing the rare class, then an LLM layer that writes the risk narrative an analyst would otherwise assemble by hand.
    Result
    AUC up from 0.60 to 0.978 and 84% recall held at that imbalance, with every flag arriving already explained.
    Next
    Tune the threshold against the real cost of a blocked card versus a missed fraud, rather than against the F1 score.
    PythonXGBoostMySQLGemini API
  • AUCKLAND · CITYQUEENSTOWN · RESORTR² 0.73R² 0.77the gap
    HospitalityMachine learning

    Airbnb price prediction across two markets

    0.77R² Queenstown
    0.73R² Auckland
    4,987Listings
    21Engineered features
    Problem
    Hosts guess at nightly price, and a city market behaves nothing like a resort town, so one model for both is worse than none.
    Approach
    3,657 Auckland and 1,330 Queenstown listings modelled separately, with 21 features engineered from review text alone, then a self-serve tool that ranks upgrades by predicted dollar impact.
    Result
    R² of 0.77 in Queenstown and 0.73 in Auckland, and a host can see which change pays before making it.
    Next
    Add seasonality, which matters far more in a ski town than it does in a city.
    PythonXGBoostNLP / LDA
  • 30 DAYS AFTER DISCHARGEcame back73%FLAGGED IN TIME
    HealthcareHealthcare ML

    Predicting hospital readmissions

    73%Recall
    25,000Patient records
    3Model families
    30Day window
    Problem
    25,000 patient records, where a false positive costs a phone call and a false negative costs someone's recovery.
    Approach
    Three model families compared, then the real work: moving the decision threshold until the error the hospital could live with was the error it made.
    Result
    73% of the patients who would return were flagged before they did.
    Next
    Price the two errors explicitly with the care team instead of inferring the trade-off from the curve.
    Pythonscikit-learnNeural nets
  • HUB 1HUB 2HUB 3
    LogisticsSupply chain

    Amazon Prime Air drone hub network

    1,900Packages / day
    3Hubs
    12 moRollout
    Problem
    Where do three drone hubs go in Massachusetts so that 1,900 packages a day actually land?
    Approach
    Sites chosen on demographics, airspace and feasibility, then a 12-month rollout with the inventory strategy and S&OP alignment to support it.
    Result
    A three-hub network with the coverage, the sequence and the stock plan to stand it up.
    Next
    Stress the siting against weather downtime, which no demographic layer captures.
    GISDemand modelingS&OP
  • SEGMENT 01SEGMENT 02SEGMENT 0312,004 PASSENGERS · PCA + K-MEANS
    TravelMarketing analytics

    What actually drives passenger satisfaction

    12,004Passenger records
    3Segments
    0.40Regression R²
    Problem
    12,004 passenger records, and an airline treating them as one audience when the average passenger does not exist.
    Approach
    Chi-square, ANOVA and regression to find which attributes actually move satisfaction, then PCA and k-means to split the cabin into segments that behave differently.
    Result
    Three segments with distinct drivers, so a fix aimed at one no longer gets judged on the others.
    Next
    Re-run it per route, since a short hop and a long haul are not the same product.
    RANOVAPCAk-means
  • 50K+RECORDS38%FASTER12KPISTHE GAP
    RetailBusiness intelligence

    Retail performance dashboards

    50,000+Records
    38%Faster decisions
    12KPIs tracked
    Problem
    50,000+ retail records that nobody had the time to read, and a weekly meeting that ran on opinion.
    Approach
    Tableau dashboards built on pre-attentive design, so the eye lands on the problem category before the brain reads a single label.
    Result
    Decisions reached 38% faster, because the gap is the first thing you see rather than the last.
    Next
    Put the definitions next to the KPIs, so nobody has to ask what counts as a sale.
    TableauKPI designData viz
Timeline

How I got here

2019

Started a BBA in International Business at Pune University.

2021

Digital marketing intern at UniText: my first taste of testing and measuring. +28% followers, +15% click-through.

2021

Joined Global Network, International Trade Advisory: three delegations, 80+ delegates, 200+ contacts in the CRM.

2023

Manager, International Markets & Operations. Led a 22-member Indian business delegation to Myanmar, hosted by the Embassy of India: 45+ B2B meetings.

2024

Moved to the US for an MS in Business Analytics at UMass Amherst's Isenberg School of Management.

2026

Graduated in May with a 3.9 GPA.

Now

AI & Data Analyst intern at Supply Bistro, Boston: sales, inventory and vendor analytics, dashboards, and forecasting.

Madhura beside the University of Massachusetts sign in 2024
Week one at UMass, 2024
Madhura on the commencement screen with her name displayed
Commencement, 2026
2021 – 2024

The trade years

Six markets where I ran the market research, coordinated the delegations and set up the B2B meetings. It all arrived as carrier records, customs paperwork and delegate lists nobody trusted. Cleaning and reconciling that is how I ended up in analytics.

PUNE, INDIAYANGON, MYANMARDELEGATION · 2023DUBAI, UAEGLOBAL EXPO · 2021–2023FRANCEUSASOUTH KOREA

USA · France · Dubai · Pune · Yangon · South Korea

  • Accompanied the delegationYANGON, MYANMAR · 2023

    A 22-member delegation, and 45+ B2B meetings run on the ground.

    The one market I was in the room for, not reading about afterwards.
  • Global ExpoDUBAI, UAE · 2021–2023

    Ran the Dubai delegations end to end: exhibitor logistics, buyer lists and the follow-up that actually got answered.

    Every edition delivered from the Pune desk, without ever being on the floor.
  • Market researchFRANCE · USA · SOUTH KOREA

    Sized the demand, shortlisted importers and set up one-to-one meetings with stakeholders for our delegates.

    Three markets screened so delegates knew who was worth meeting before they travelled.
4+Multi-sector
delegations
120+Delegates
coordinated
45+B2B meetings
in Myanmar alone
1,200+Records reconciled
across markets
Experience

Where I've worked

Aug 2026 – NowBoston, MA

AI & Data Analyst (Intern)

Supply Bistro
  • Analyse sales, inventory and vendor data in SQL and Excel to surface usage trends and cost-saving opportunities.
  • Build and maintain Tableau and Power BI dashboards for sales, inventory and vendor performance.
  • Develop AI-driven automation for data extraction and recurring reporting; apply predictive modeling to inventory planning.
Oct 2024 – May 2026Amherst, MA

Management Student Supervisor

UMass Auxiliary Enterprises · Part-time
  • Coordinated day-to-day service operations across multiple units, resolving workflow delays to keep service moving through peak hours.
  • Studied inventory movement patterns to guide replenishment decisions, reducing shortages during high-demand periods.
  • Reviewed workload distribution to support shift planning and staffing, alongside cashier duties and front-line guest service in a high-volume setting.
Jan 2023 – Aug 2024India

Manager, International Markets & Operations

Global Network · International Trade Advisory
  • Drove a 39% B2B conversion improvement across 4+ multi-sector delegations (120+ delegates) through tests, hypotheses and executive briefings.
  • Resolved anomalies across 1,200+ records, lifting accuracy 30% and enabling risk-informed S&OP decisions.
  • Built the metrics and visualisations senior executives worked from (28% faster decisions); cut project lead time 35% across 60+ stakeholders.
Dec 2021 – May 2022India

International Marketing Executive (Intern)

Global Network · International Trade Advisory
  • Organised 3 multi-sector delegations for 80+ delegates at 95% attendance.
  • Researched 5+ international markets; managed CRM for 200+ contacts; 40+ B2B meetings at 70% conversion.
Feb – May 2021India

Digital Marketing (Intern)

UniText
  • 28% follower growth and 15% CTR improvement across 3 channels; 45+ posts and 8+ articles lifted engagement 32%.

Programming & data

PythonSQLRMySQLAPIsAdvanced Excel

Analytics & BI

TableauPower BIKPI designDashboardsData storytelling

Modeling

scikit-learnXGBoostForecastingHypothesis testingA/B testing

Cloud & automation

AWS S3LambdaCloudFrontAlteryxUiPath

Certifications

GenAI & LLMs · MicrosoftPower BI Fundamentals · DataCampStatistics for Data Science · UdemyAWS Jam · Cloud Security & AutomationAWS Cloud Practitioner (in progress)Introduction to RMySQL BasicsFull Stack Python
Showing up

In the room

Industry visits and events: where I go to hear how analytics actually gets used, and to ask too many questions.

Madhura at the Microsoft office logo wall in Burlington
Microsoft · Burlington · 2025
Madhura at Gillette Stadium on an Isenberg School visit
Gillette Stadium · Foxborough · 2026
Group photo from the Finance Trek in New York City
Finance Trek · New York City · 2024
Madhura at the AWS Summit in New York City
AWS Summit · New York City · 2026
Hobbies

Off the clock

Madhura skiing on a snowy slope
On skis
A graphite pencil portrait drawn by Madhura
Sketching, my oldest hobby
Madhura at a mountain viewpoint above a lake
Worth the climb
Madhura's potted plants in the sun on her balcony
Outgrew the windowsill
Madhura cooking large pots of rice in the kitchen
Happiest when I cook
Madhura walking down a flower-lined city street under palm trees
I love a city walk
Contact

Got a messy
dataset?

Let's talk.

Open to analytics, data science and BI roles · Based in Boston, open to relocation.