Apr 2026 – Jun 2026
Data Analyst (Capstone Project)
BC Children’s Hospital · UBC MDS Capstone
Investigating the impacts of Computerized Provider Order Entry (CPOE) on Pediatric Laboratory Testing. Full time Capstone Project for the UBC MDS program.
Athul Sasidharan
Mechanical engineer, Business Consultant, then data engineer, now a data scientist in Vancouver. I’m best where the data is messy, the deadline is real, and somebody needs an answer they can defend afterwards.
Education
Sep 2024 – May 2025
Global Leadership Graduate Certificate
2018 – 2022
Bachelor of Technology, Mechanical Engineering
Work experience
Jul 2022 – Feb 2024
Data Engineer
Axis Bank Ltd · Bangalore
- Built ingestion and transformation pipelines in Oracle, PL/SQL and Informatica serving five-plus reporting teams, under banking regulatory constraints where a wrong number is a filing problem rather than a dashboard problem.
- Owned unit testing and production release in Blaze/TRAID, catching breakages pre-deploy to keep live environments stable.
- Translated business requirements into data models with stakeholders in the room, then wrote the documentation standard the team adopted.
May 2020 – Apr 2022
Business Consultant
Part-time · Agies Safety Solutions · Kerala
- Designed a new product sales model from statistical analysis of the existing book, contributing to 5× revenue growth.
- Automated finance and administrative reporting, removing up to 30% of annual operational turnover in avoidable cost.
- Ran two years of market and competitor research that leadership used to set direction.
Jun 2025 – present
Cash Supervisor
Shoppers Drug Mart · Vancouver
- Register, lottery terminal, merchandise intake, backroom inventory and end-of-day close in a high-volume pharmacy, alongside the master’s.
- Logistics, budgets, sponsor negotiation and on-site escalation for 75+ events across 14 departments and 13,000+ participants.
Projects
Jul 2026 · HCLTech × UBC hackathon
CodeIQ — codebase knowledge graph
Parses a 100K-line React Native codebase into a queryable knowledge graph, then answers plain-English questions about structure and impact. Confidence is scored deterministically from retrieval relevance rather than self-reported by the model.
whyEvery unfamiliar codebase starts with two hours of “wait, who calls this?” We wanted that down to one question.
28/30 on the benchmark set, zero errors2026 · Python package
geospatial-toolkit
A lightweight package standardising the geospatial operations people rewrite by hand — coordinate formats, great-circle distance, point-in-polygon city lookup, and antipode calculation with reverse geocoding, which nothing else offers in one call.
whyHalf the cohort was rewriting the same distance function. We packaged it so the other half wouldn’t have to.
v3.0.0 tested, documented, published to TestPyPIApr 2026 · Retrieval system
Smart product review search
Hybrid retrieval over Amazon software reviews — BM25 and dense embeddings fused with reciprocal rank fusion, then an LLM that has to cite the review it used. Benchmarked against a faster model and kept the slower one, because it admitted retrieval gaps instead of over-committing.
why16,000 reviews and no way to ask them anything. Now it either shows its receipts or keeps quiet.
15,972 reviews indexed and source-cited2026 · Deployed dashboard
Vancouver Park Finder
A public dashboard for searching Vancouver parks against whatever a person actually cares about — washrooms, dog areas, playgrounds, distance from home.
why248 parks in this city and no simple way to ask which one has a washroom and lets the dog off the leash.
248 parks, filterable in one screen2026 · Cloud deployment
Rainfall prediction API
A model trained on an EMR/Spark cluster, packaged behind FastAPI and served from a t3a.micro EC2 instance. Loaded once at import rather than per request, with Pydantic-validated payloads and cross-testing against a classmate’s endpoint.
whyA model sitting in a .joblib file has never helped anybody. This one at least answers the phone.
2026 · Reproducible ML
Hotel cancellation prediction
A classification study on hotel booking cancellations, built so a stranger could re-run it: pinned environment, containerised, scripted data flow, and a report that regenerates from source.
whyRan it, got a number, couldn’t reproduce the number a week later. Built this so that stops happening.
Built because I wanted to
Jan 2026 · Writing
Multi-armed bandits, explained
A written walkthrough of the explore-versus-exploit problem and why A/B testing is often the wrong tool for the job it gets used for.
whySlot machines have taught statisticians more than statisticians have taught slot machines. Wrote it down before I forgot why.
Apr 2026 · Tooling
NotebookLM on Claude Code
An integration bridging Google NotebookLM into Claude Code, with automated version management, API safety checks and rollback when a change misbehaves.
whyTwo tools I use constantly that refused to speak to each other. Now they do, and there’s an undo button.
Toolkit
Query & language
- SQL
- Python
- R
- PL/SQL
- Bash
Analysis & modelling
- pandas
- scikit-learn
- dbplyr
- tidyverse
- Causal inference
- Experiment design
Pipelines & cloud
- Informatica
- Spark
- AWS EMR / S3 / EC2
- Warehouse design
- Oracle
- PostgreSQL
- SQL Server
Reporting & delivery
- Power BI
- Shiny
- Dash
- Quarto
- ggplot2
- Streamlit
LLM systems
- RAG
- BM25 + dense hybrid
- FAISS
- ChromaDB
- LangChain
- Evaluation harnesses
Practice
- Git
- Docker
- CI
- Reproducible research
- Data science ethics