Experience in industry
Scribd, The world’s largest digital library
Scribd is a digital reading subscription service, containing over 1M eBooks, audiobooks and magazines.
Role: Applied Research Data Scientist (2019 — present)
Applied research in natural language processing and deep learning.
- Named Entity Detection and Entity Linking
- Fine tuning language models for sentence classification
- Topic analysis
- Long form document classification
- Time series clustering
- Developed graph embedding models
ROSS Intelligence, Legal technology
ROSS is a legal research service leveraging advanced natural language processing.
Role: Machine Learning Engineer (2017 — 2019)
- Learning-to-rank: Developed a search engine to find the most relevant case law to a user-provided question using Deep Learning and NLP.
- Learning-to-rank: Developed features for the ranking algorithm based on latest LtR literature.
- Natural Language Processing: Developed text processing pipeline.
- Information retrieval: Reduced the retriever latency by over an order of magnitude (>10x) by controlling the shard sizes and respecification of AWS resources.
- Built a memoization service that further reduced search latency by 60%.
- Built a platform for experimentation with information retrieval.
OR Empreendimentos, Real Estate Investment
OR is one of the largest real estate developers in Brazil in the residential, business, commercial and tourism markets, with over 6B USD in assets on the market between 2008 and 2016.
Role: Financial Analyst (2014 — 2016)
- Automated several workflows in the department, allowing reports to be generated daily instead of monthly. This drastically improved management reaction time and saved an estimated 120 man-hours per month.
- Created a daily report for bank balances of over 300 regional bank accounts, allowing excess amounts to be invested, yielding over 3M USD monthly.
- Redesigned the budget model, giving management fine control of over 100 regional offices’ five month financial plans. Savings estimate not disclosed.
- Modeled bank risk. Not all banks have Credit Default Swaps (CDS) instruments on the market, which makes risk assessment difficult. The solution was to use Libor spreads to synthesize CDS values.
The final model corroborated the previous model’s results. - Modeled arbitrage with excess funds and the cost of cash.
- Organized monthly department meetings to evaluate company metrics; politics and macroeconomics and their potential effects on the real estate market; new monthly objectives.
Zocprint, Rocket Internet Venture
Zocprint is the #1 e-commerce of customizable printed marketing materials and gifts in Brazil.
Role: Head of Customer Service (2013)
- Developed the Customer Service department from scratch, defining business processes, data generation processes and key performance indicators.
- Hired and managed 9 customer service agents who had individual productivity and quality measurements as well as weekly goals.
- Closed the quality feedback loop to the Operations department by automatic summarization of product issues.
- Browsing pattern analysis to reduce demand for customer service.
Kondor Invest, Hedge Fund
Kondor is an asset management firm with a rigorous proprietary risk management solution.
Role: Risk management intern (July-December 2012)
- Responsible for ensuring all daily trades were correctly booked.
- Development of an e-mail parsing tool to verify correct trade booking.
- Introduced double checking of data entry as an operational risk mitigation strategy.