Pallavi Bhimte
- Data Scientist
- AI Engineer
- ML Engineer
- Data Analyst
- Data Scientist
Three years in consulting taught me to build data and AI systems that quietly do the hard work — from messy problem to something people actually rely on.
Selected work
Client projects at Deloitte and Emmi. Scoped, built, handed over.
Workers' Compensation Chatbot
An end-to-end AWS Lex conversational system covering 90+ query types across 95 flows. I defined a 12-scenario evaluation framework and led six SMEs through model training, then demoed it live to 13 senior executives. Adopted organisation-wide after the pilot.
Cloud Data Platform
Architected a four-pipeline AWS + dbt platform and migrated 500+ SAP extractors onto it. Worked across nine client stakeholders in product, BI and business, then handed the whole thing to a five-person implementation team with the documentation to run it.
SAS Migration Assessment
Assessed 120+ SAS scripts across nine systems and modelled three migration strategies against value, effort and risk. Presented a three-phase roadmap with four options to senior stakeholders, sized at 5–7 person-months.
Portfolio Carbon Analytics
Built a Python application that quantifies and explains the carbon emissions inside a customer's investment portfolio — turning dense financial data into something a product team could actually reason about. Designed the Snowflake tables underneath it for KPI tracking and reporting.
Things I built on my own
Personal projects, open-source repos — click through and read it.
ResearchMind
Four specialised agents that research a topic together: a Search agent hits Tavily, a Reader agent scrapes the good sources, a Writer chain drafts the report, and a Critic chain scores it and lists what's weak. The critic is the interesting part — it's honest enough to give its own pipeline a 6/10.
View repoAdvanced RAG
A worked series on retrieval, built up from scratch: simple RAG, then embedding models, then semantic chunking, then contextual retrieval. Includes t-SNE projections of the token embeddings and input/output similarity heatmaps, because the failure modes are much easier to see than to describe.
View repoFastAPI + Model Serving
Two APIs. One serves a Random Forest car-price model — loaded once at startup, one-hot encoding handled server-side, with a Streamlit front end to poke at it. The other is an e-commerce REST API built to push Pydantic hard: nested models, computed fields, custom validators, allowlisted seller domains.
View repoThe path
Journey through five industries, two continents, and a lot of data.
Experience
Consulting taught me the hard part is never the model.
-
Emmi
Dec 2024 — Feb 2025 MelbourneData & ML Engineer
- Built an analytics-driven Python application quantifying and explaining customer carbon emissions from investment portfolios, translating complex financial data into interpretable insights for product and business teams.
- Designed analytics-ready Snowflake tables supporting KPI tracking, dashboards and product reporting.
-
Deloitte
Jan 2022 — Nov 2024 MelbourneAI & Data Consultant
- Architected an AWS + dbt cloud data platform ingesting 450K+ records/day across 42 sources; migrated 500+ SAP extractors and cut manual reporting effort 70%.
- Built an AWS Lex conversational AI system covering 90+ query types at ~90% production accuracy, reducing routine support query volume 70%; adopted organisation-wide after pilot.
- Assessed 120+ SAS scripts across nine systems, projecting 30% cost and 60% complexity reduction in a three-phase migration roadmap.
- Built PySpark workflows processing 100K+ member records across three release cycles — 98%+ data accuracy, 70% less manual reconciliation.
- Shipped an LLM-powered internal automation app (OpenAI API) standardising seven job task categories across two teams.
Stack
Shipped production work with most of it. Hover the ticker up top to read it properly.
Languages
- Python
- SQL
- PySpark
- R
GenAI
- LangChain / LangGraph
- Azure OpenAI
- Pinecone / RAG
- FastAPI
- LangSmith / CrewAI
Cloud & data
- AWS
- Databricks
- Snowflake
Tools
- Airflow
- Terraform
- dbt
- GitHub
ML & analytics
- scikit-learn
- Pandas
- NumPy
- A/B & hypothesis testing
About
Drop
pallavi.jpg next to index.html and replace this div with an img tag. 4:5 portrait works best.
I moved from India to Melbourne in 2020 for a Master's and stayed.
Most of what I know came from being handed an ambiguous problem, a room of stakeholders who didn't agree with each other, and a date. Three years of consulting will do that. The model is rarely the hard part — the hard part is figuring out what people are actually asking for, and then getting the thing into production without it quietly rotting.
What I like is the moment something starts working. A chatbot that takes 70% of the tickets so nobody has to answer them. A pipeline that runs at 3am without waking anyone. A report that changes someone's mind in a meeting.
Right now I'm deep in agentic systems — building them, breaking them, and working out what actually survives outside a demo.
- Currently building
- What's on the bench right now
- Currently learning
- Something specific
- Off the clock
- The non-work answer
- Based in
- Melbourne, VIC
Credentials
Five certifications across AWS and Databricks. Every badge links to its issuer verification page.
-
AWS Solutions Architect
Associate
-
AWS Cloud Practitioner
Foundational
-
Databricks Data Engineer
Associate
-
Databricks Generative AI
Fundamentals
-
Databricks Lakehouse
Fundamentals
Education
-
Master of Data ScienceRMIT University, Melbourne — Mar 2020 to Nov 2021 -
B.Tech, Computer Science & EngineeringGalgotias University, India — Aug 2015 to Dec 2019
Let's talk about your messy data.
Open to Data Scientist, AI Engineer, ML Engineer and Data Analyst roles in Melbourne or remote. Australian permanent resident — full working rights, no sponsorship needed.
