Email: puneet [dot] ludu [at] gmail [dot] com
Location: New York, NY
Phone: +1-(716) eight six seven four three four four
Website: puneet.io
Zillow (Zestimate), Remote
Sep 2021 -- Present
- Architected and led the end-to-end development of an interactive Comparative Market Analysis (CMA) platform with Realtime Valuations, Property Embeddings and Comps API.
- Impact: 0 to 1 project to boost engagement and satisfaction, paving the way for new revenue streams.
Zestimate Infrastructure Modernization (Python, Terraform, AWS, Kubeflow, Metaflow, Docker, Gitlab CI)
- Led the modernization of a critical valuation ML infrastructure, transitioning to more cost-effective, containerized technologies.
- Impact: Achieved operational improvements and annual cost savings of $500k.
- Integrated advanced machine learning tools into team workflows and established coding standards.
- Impact: Improved overall team efficiency, code quality. Reduced On-Call alerts by 95%.
- Managed interns and Mentored new hires and junior engineers, fostering technical skill development and guiding them through project contributions.
OkCupid (Match.com), New York City
May 2020 - Sep 2021
- Lead the efforts to optimize subscription pricing(discounts) to maximize the revenue for OKCupid.
- Impact: Increased overall revenue by 6% through A/B testing against assigned prices.
FactSet, New York City
Apr 2015 - May 2020
- Developed and deployed an end-to-end speaker identification system to identify speakers in real-time during company quarterly earnings calls using computer vision and deep neural networks.
- Impact: In early testing it was estimated to save around 20% human-hours.
- Lead the efforts to extract 'full company name' with key-people, their titles and biographies etc. from 1.6 million crawled and cached websites of private companies.
- Developed full-stack solution to identify the duplicate documents in real time, given a stream of thousands of documents per day.
- Impact: 66% reduction in compute time for document processing. Also, used by StreetAccount to find trending news.
- Lead developer for implementing features like Autocomplete Query(Type Ahead) and suggest similar concepts to expand the formulated query for a 'Financial Document Search Engine'.
- Developed the pipeline to cluster users and rank the formulas in the feature of FactSet terminal.
- Impact: Average rank brought down from 5.6(ElasticSearch based) to 2.3(Language Model based).
Tata Research Development and Design Centre, India
July 2011 - July 2013
- Wrote an algorithm based on Shape Context for finding frequently occurring patterns and events, with as good results as SAX, DTW etc. with 7% better results in the particular domain of car sensors.
- Implemented an ETL framework that exploits the power of map-reduce and big-databases to fuse incongruous enterprise data from disparate sources in near real time.
Languages: Python, Java, C/C++, Bash, Javascript, HTML, SQL
Frameworks: PySpark, Keras, Metaflow, KubeFlow, TensorFlow, PyTorch, MongoDB, FastAPI, Django
- Inferring Latent Attributes of an Indian Twitter user using Celebrities and Class Influencers, ACM Hypertext 2015
- Inferring gender of a Twitter user using celebrities it follows, CORR 2014
- Architecture for Automated Tagging and Clustering of Song Files According to Mood, IJCSI, 2010
Master of Science in Computer Science, State University of New York, Buffalo, NY
B. Tech. in Computer Science and Engineering, JIIT, India
- Organizer @ MUFin: Committee member, organizer and reviewer to the MUFin Workshop at top conferences, focusing on innovative approaches to modeling uncertainty in the financial sector (AAAI2023, PKDD2022).
- Lotion: Unofficial Notion.so Desktop app for Linux (2K+ GitHub stars / 60K+ Clones & Downloads).
- Romadeva: Tool to convert Roman script to Indic(Devanagari) script (Used by https://translatorswithoutborders.org).
- jTextBrew: A JAVA library for fuzzy string matching, based on TextBrew algorithm by Chris Brew.
- Quena: Question and Answering system -- Indexed 1.6 Million Wikipedia documents, designed a question parser and a ranking algorithm based on popularity. (Apache Solr, NER, POS tagger).
All previous versions are available in the versions directory.
For instructions on how to compile this resume, see RUNSTEPS.md.
