Shivani's Desktop

about_me.txt - Notepad

SHIVANI SRIVASTAVA
Senior Data Scientist, New Delhi, India

I am a natural resource economist by training, and data scientist by profession.

What I think I do particularly well is connect things that were not originally designed to be connected: methods from different fields, seemingly unrelated datasets, and ways of framing problems borrowed from disciplines far outside the one I started in. I like taking those pieces apart, finding the useful common thread, and putting them back together into a solution that is both rigorous and unconventional.

I am also a massive admirer of Taylor Swift and Agatha Christie. Consequently, there are several Easter eggs (some underway), clues, and questionable decisions scattered throughout this website. Finding them is optional. I will, however, be quietly impressed if you do.

As of today, I am open to full-time opportunities where I can work at the intersection of unconventional data, rigorous analysis, and creative problem-solving —particularly problems where the obvious method isn't necessarily the interesting one.

Focus

ML · NLP · Generative AI · Decision Intelligence

Reach me

shivani.srvstv.work@gmail.com · github.com/Shivani-Srivastava

C:\projects

I collect strange questions and try to answer them with data.

The projects below are less about the tool I used and more about the connection I made: a problem, an unusual signal, a method borrowed from somewhere else, and what came out of putting them together.

PDF

Can Social Media tell us what a park is worth?

Using Crowd-Sourced Data on a Travel Cost Model to Approximate Value of Recreation in Urban Green Spaces: Lodhi Garden Case Study

ECONOMICS SOCIAL MEDIA ENVIRONMENTAL VALUATION

I combined a classical environmental-economics valuation method with social-media data to estimate something that isn't directly observable: the recreational value people derive from an urban green space.

WHY THIS?

I wondered whether people were already leaving behavioral traces of recreation online, and whether those traces could be connected to a classical Travel Cost Model.

METHOD

Geo-tagged social media visitation information plus travel-cost modeling to approximate recreational value for Lodhi Garden.

M.Sc. Thesis · TERI SAS · 2021

DB

Six datasets. Zero primary keys.

Building an Entity Resolution Engine to Map Disparate Power-Plant Databases

ENTITY RESOLUTION DATA ENGINEERING RECORD LINKAGE

I built a process to identify and link entities across six datasets despite missing identifiers and incomplete data, including an efficient manual-validation layer.

WHY THIS?

Real-world datasets rarely arrive with the clean identifiers that textbook examples assume. The challenge was to create a reliable canonical view without pretending the missing information existed.

METHOD

Entity matching across six sources plus candidate generation and efficient manual validation for ambiguous records.

Genpact · 2024–2026

TXT

Can an earnings call reveal how much control a company thinks it has?

Using Earnings Call Transcripts and LLMs to Identify Organizational Locus of Control

LLMs NLP PSYCHOLOGY FINANCE

I used executive language in earnings calls as a behavioral signal for an otherwise abstract psychological construct: organizational locus of control.

WHY THIS?

Locus of control is psychological, while earnings calls are financial communications. The project asks whether language produced in the latter can act as evidence for the former.

METHOD

Earnings-call transcripts plus LLM-assisted language analysis to identify indicators associated with organizational locus of control.

ISB · Research project

EYE

Where do people actually look?

Using Eye-Tracking Surveys to Identify Areas of Interest to Viewers, Comparing Content Creator and Brand Language

EYE TRACKING EXPERIMENTAL DESIGN MARKETING

Instead of asking people what caught their attention, I measured where their attention actually went and examined how creator versus brand language affected it.

WHY THIS?

Stated attention and observed attention are not necessarily the same thing. Eye tracking provides a behavioral measure rather than relying entirely on what viewers say they noticed.

METHOD

Eye-tracking survey data plus areas-of-interest analysis and comparison of creator and brand language.

ISB · Behavioral research

BAY

What stereotypes do we form without realizing it?

Bayesian Conjoints: Identifying Visual Stereotypes Formed by Review Consumption

BAYESIAN METHODS CONJOINTS VISUAL PERCEPTION

I used Bayesian conjoint methods to investigate how exposure to review content can shape visual associations and stereotypes.

WHY THIS?

The interesting question wasn't simply whether people preferred something, but whether repeated information consumption changes the visual associations they form.

METHOD

Conjoint-style experimental design plus Bayesian estimation to examine preference and stereotype formation.

ISB · Behavioral research

ML

When a prediction has to become a decision.

Customer Segmentation → Tiered ML Models → Customized Contact Strategies

MACHINE LEARNING SEGMENTATION DECISION INTELLIGENCE

I combined K-Modes segmentation with tiered ML models to move from identifying customer groups to determining how those groups should actually be approached.

WHY THIS?

A model is only useful if its output changes what someone does. The objective was therefore not just prediction, but translating model outputs into differentiated contact strategies.

METHOD

K-Modes customer segmentation plus tiered risk/propensity models and customized contact-strategy recommendations.

Genpact · 2024–2026

AQI

Did a pollution policy actually work, or did the weather just get lucky?

Using Satellite Data and Synthetic Control Methods to Evaluate Delhi's Graded Response Action Plan (GRAP)

CAUSAL INFERENCE GEOSPATIAL PUBLIC POLICY

I built an early-warning dashboard for air pollution across Delhi NCR and used a Synthetic Control approach to test whether Delhi's GRAP policy actually reduced pollution, or whether the observed drop could be explained by something else entirely.

WHY THIS?

Policy evaluations often conflate correlation with causation: pollution can drop for reasons that have nothing to do with the policy itself, like weather or a slowing economy. The goal was to separate the policy's actual effect from coincidence.

METHOD

Satellite and geospatial pollution data processed via Google Earth Engine, feeding an early-warning dashboard, plus Synthetic Control modeling to construct a counterfactual Delhi without GRAP.

CEEW · Feb–Mar 2024

career.exe — Properties

Trajectory

Genpact
May 2024 – Jul 2026 · Gurugram, India Business Analyst → Data Scientist → Data Engineer

Research Intern · CEEW
Feb – Mar 2024 · New Delhi, India

Research Assistant · ISB
Sep 2021 – Jan 2024 · Hyderabad, India

Stack

Python · R · SQL · Scikit-learn · XGBoost · Pandas · Streamlit · Shiny · Power BI · Tableau · Google Earth Engine · Knowledge Graphs · AWS

Leadership

Leads a team of 5 Data Scientists — Genpact

Mentored 12 interns — ISB

education.txt - Notepad

Degrees

PG Diploma, Data Science — IIIT Delhi (2024–25)

M.Sc. Environmental and Natural Resource Economics — TERI SAS (2019–21)

B.A. (Hons.) Economics — Banaras Hindu University (2016–19)

Recycle Bin — cv-of-failures.pdf

Not everything makes the highlight reel. This is the rest of it.

If the preview above is blank (common on mobile browsers), open the PDF directly.

Alert

!

Work in Progress

This section isn't ready yet — check back soon.