Ana Areias

I’m Ana Areias, a Development Economist and Data Scientist.

Controlling my cursor with my fingertip using handtracking with MediaPipe Hands

Today, humans generate exponentially more data than ever before, about behaviors or processes that used to be unmeasurable, in forms that previously would never before have been considered “data.”

I get a thrill from looking at the world through the ultimately very creative lens of a data scientist—what is “data-izable,” or “data-fiable,” that could be mined for new insights about human behavior? How can I combine previously disparate data to generate predictions that will help improve business processes, or even human quality of life?

I love the fact that the job can be a combination of quant-y wizardry, visual communication, great storytelling and business savvy.

I’m currently a data scientist at Kineviz. I work on merging deep learning and graph technologies in GraphXR, our graph data BI and visualization platform, to better understand high dimensional and interconnected data.

Previously I’ve worked on poverty prediction and big data for labor at the World Bank. I’m a Harvard MPA-ID grad and have been a Fellow at Data Incubator and member of DataCorps with DataKind. I’ve also been a Program Manager at Data-Pop Alliance.

When I’m not coding for fun I’m an avid surfer and a struggling billiards player.

Have a look at some projects I’ve worked on with Kineviz:

• Fine-tuning to enhance NLP tasks with small, self-hosted language models - Instruction fine-tuning Mistral:Instruct 7B improved standard named entity recognition (NER) F1 score by 12% and domain-specific NER by 30%.

• Locally hosted LLMs for entity and relationship extraction - Comparing the quality of responses generated with self-hosted Mistral-7b with OpenAI’s ChatGPT

• FluBERTa - UMAP projection of RoBERTa Transformer embeddings learnt from 33k Flu sequences shows distinct hemagglutinin groups within host organism groups. Part of Kineviz’ Art of Graph series and animated visualization.

• NODES 2021 - Neo4j Conference Presentation in Data Science track video available on conference page and slide deck

• DreamCatcher: Deep Learning and Graph Analytics for the Dreamscape blogpost and companion Jupyter notebook

• Twitter Demographics and the Election 2020

• DJDC #34 - Twitter’s #Elections2020 through the lens of data science + the arts

with the World Bank;

• Machine Learning and Prediction of Beneficiary Eligibility for Social Protection Programs. Chapter 8 in book “Revisiting Targeting in Social Assistance : A New Look at Old Dilemmas”

• Large Number of Job Ads in India Explicitly Prefer Men: World Bank Study

• Asymmetric Information on Noncognitive Skills in the Indian Labor Market: An Experiment in Online Job Portal

• Toward labor market policy 2.0 : the potential for using online job-portal big data to inform labor market policies in India

with DataKind;

• Using Data to Drive Donations: key findings from our work with DataKind

• The Fascinating Day Jobs Of DataKind Volunteers

• Donations by Data

• Stories from the Frontlines of Data Science for Good

and for fun:

• Link Prediction: Predicting Author Collaborations with Neo4j, Spark MLlib and Node2Vec

• Dollar Blues: Predicting the Black Market Exchange Rate in Argentina through Google Search Behavior

• Automatically update your wallpaper with Magicseaweed’s Photo of The Day

• Shell script for Pomodoro timer and task logger

• Sailing Club of Washington (SCOW) by the numbers

For more details and contact info check out my Linkedin.