(01) · learning-to-rank
Expedia Hotel Booking Analytics
Reordering hotel search results so the one you book lands on top. 4.96M records, 146 features, top 15% of 110 teams.
msc business analytics · vu amsterdam
Shreya Bhatt. I turn messy data into answers.
I'm Shreya, an MSc Business Analytics student at VU Amsterdam with three years of fintech data work behind me at Galytix, where I built the pipelines that turned unstructured financial reports into analyst-ready data for Société Générale.
Python, SQL, ETL pipelines, data modelling, OCR extraction, LightGBM, learning to rank, recommendation systems, PyTorch, statistical modelling, counterfactual simulation, Streamlit, AWS
In numbers
3 yrs
fintech data work at Galytix
1,200+
global companies in the datasets I built
top 15%
of 110 teams on a learning-to-rank task
1.9M
Amsterdam City Card transactions analysed
(the short version)
I've read more financial statements than most people ever will. For three years I turned scanned PDFs and financial tables in mixed layouts into one clean schema that credit analysts could actually compare. Now I'm in Amsterdam, building models that rank and recommend, and caring just as much about whether the result is useful to a real person.
more about me →how i work with data
From messy input to something people use.
The part I enjoy is the whole trip: getting raw data into shape, working out what it actually says, and handing over something a person can act on.
(01 · in)
Get it clean
Messy inputs become data people can trust.
- Turned scanned PDFs and mixed-layout financial tables into one standardised schema
- Normalisation logic that reconciled reporting formats across institutions and fiscal years
- Automated OCR extraction: two hours saved per day, ~20% more throughput at peak
(02 · through)
Make it mean something
Features, models and validation that hold up honestly.
- 146 features engineered from 54 raw columns across 4.96M Expedia records
- Learning-to-rank: 16th of 110 teams, NDCG@5 of 0.418, with temporal validation
- Context-aware recommender for Amsterdam tourist flows
(03 · out)
Put it in front of people
Results someone can act on, not just a longer slide deck.
- Analyst-ready datasets for credit analysts at Société Générale, covering 1,200+ companies
- Counterfactual simulation: 3,617 fewer visits at peak locations
- Interactive Streamlit dashboard for crowd-management review
selected work
Projects
(02) · recommendation system
Tourist Flow Optimisation
Nudging Amsterdam visitors away from the crowded spots. 1.9M City Card transactions, a context-aware recommender, and a counterfactual simulation to measure the effect.