Data engineering · Machine learning · Automation
We build the pipelines that keep your data clean, the models that turn it into predictions, and the automations that save your team 20+ hours a week. From everyday data management to complex modeling, with deep experience in data on how people behave: customers, members, donors, and voters.
What we do
ETL pipelines, API integrations, web scraping, and record matching, with validation checks built in so problems surface early instead of in your reports.
Predictive models for classification, forecasting, fraud detection, churn, retention, and prioritization, built and validated in Python for real-world use.
Scheduled jobs, automated reports and documents, and LLM-powered workflows that replace hours of manual copying, cleaning, and formatting every week.
Census and demographic data, geocoding, mapping, and statistical models that show where things are happening and why.
Voter files, district boundaries, targeting models, and field reporting, from people who have run data for statewide operations.
Dashboards, charts, and maps that explain what the data means to people who don't work with it every day.
Selected work
ETL · Dashboards
Problem. A client's fundraising records were being processed by hand, with no clear view of where giving came from.
Approach. A Python and SQL pipeline loading records into PostgreSQL, feeding a live Tableau dashboard of giving across every NYC ZIP code.
Result. The manual process was retired, and the dashboard guided targeted campaigns that lifted conversion.
Python · API integration
Problem. Researchers needed reliable vaccine literature pulled from a large biomedical database and sorted by trustworthiness.
Approach. A Python tool that queries the NCBI API in batches with retry handling and scores each source against WHO Vaccine Safety Net criteria.
Result. Sources sorted into a five-tier ranking and exported in multiple formats, ready for review.
NLP · Geocoding · Census data
Problem. A research project needed to know which real places appear in popular song lyrics, and where.
Approach. Extracted place names with NLP, then geocoded and enriched them through the Geocodio, Last.fm, and Census APIs.
Result. About 60,000 place references pulled from 275,905 lyrics, mapped and modeled for regional patterns.
How we work
We read your code and data before quoting, and say up front where effort is hard to predict.
A small test run comes first. If the numbers threaten the timeline or budget, you hear about it before we continue.
Records we can't validate go on a review list rather than into your dataset as a low-confidence guess.
Who we are
Cold sunlight: the clear, low winter light that shows things exactly as they are.
Co-founder · Pipelines, spatial analysis & statistics
Charles keeps data pipelines dependable and turns location and demographic data into clear answers. He has managed databases with more than 22 million records, built the automations and dashboards that large teams rely on every day, and used statistical and spatial models to decide where outreach will do the most good. He holds an M.S. in geography focused on data science and spatial analysis, and he explains results in plain language rather than jargon.
Co-founder · Machine learning & automation
Abir is a data scientist and data engineer specializing in machine learning, data infrastructure, and automation. He builds end-to-end data solutions, from ETL pipelines and API integrations to production machine learning models and reporting systems, using technologies including Python, SQL, PySpark, BigQuery, and AWS. His work has included fraud detection, retention modeling, forecasting, prioritization, and LLM-powered automation.
Tell us what you're working with. We'll reply with questions and an honest read on scope.