An independent data and AI studio in Portland, Maine

Custom data and AI tools you will understand and own.

Built for your messy problem, the one big software does not cover.

Katahdin Logic designs, builds, and teaches. Each project is made for one organization and fitted to how it already works. The careful work happens before launch, and it happens with you: we define what the tool should and should not do, try it on your real cases, adjust it until it is right, and plan how it will be maintained and upgraded over time. At launch, your team understands how it works and how it will evolve, and the code and the data are yours.

Founded by Allison Dzubak, Ph.D., a former Bowdoin College faculty member with a doctorate in computational chemistry.

Design

Before anything is built, we work out what a tool should do, what it would cost to run, and what you would get for the money. A design can be the whole job or the first step toward a build.

Example engagement

You have seen ChatGPT or Claude handle your task in a chat window. Running it automatically at scale is priced differently, and prices vary more than tenfold between models. We test a set of your real cases on several models, score every result against criteria you approve, and project each one's cost at your volume. You see what you gain and give up at each price, and can make an informed decision based on the results.

Build

Small, custom tools fitted to how your organization already works. Each one takes a job your team does by hand, or cannot do at all, and makes it routine. You own the code and the data, with no license fee and no per-seat pricing.

Example engagement

Two of your systems each hold half of an answer, and getting it takes an afternoon of spreadsheets. We build a dashboard that pulls from both automatically and applies the rules behind your bills or reports. Every figure on it traces back to its source record and the rule that produced it, so when someone questions a number, you can show exactly where it came from.

Teach

Hands-on trainings during and after a build, planned with Universal Design for Learning practices by a former college professor. Every tool comes with detailed documentation written for the people who use it.

Example engagement

Midway through a build, your team tries the tool on their own real tasks while we collect feedback that shapes the next round of the design. After launch, each role gets a session built around its own work and a step-by-step guide to keep. Everyone, technical or not, is glad to use it, because they have already tried it, asked for changes, and seen it work the way they wanted.

Featured Work

Congressional bills in plain English, every day

Legislative Transparency Platform

Live

A public bill tracker built for a civic transparency organization. Every day it collects new Congressional bills from Congress.gov and publishes a plain-English summary of each one. The AI is instructed to work only from the bill's own text, and every summary links to the original, so a reader can check it in one click.

Design

Aligned with the client on what a summary may draw on, what shape it takes, and what it leaves out.

Build

A public website that collects, summarizes, and publishes new bills every day with no manual steps.

Teach

Iterative live sessions, decks, and docs covering platform operations and admin dashboard monitoring, with client preferences built into the design.

Bill tracker showing stats banner and legislative status categories

Browse by Status

Bills grouped by where they stand in Congress, with current counts, from introduction through committee to floor votes and signing.

AI-generated bill summary with expandable details

Summaries Held to the Bill's Text

Each summary has two layers: key points at a glance, plus expandable detail for deeper reading.

Multiple bill cards showing diverse policy areas

From Congress.gov to Plain English

Raw legislative data turned into readable, searchable bill cards covering many policy areas.

The platform runs on Flask and PostgreSQL, deployed on Railway with scheduled jobs for daily bill ingestion from Congress.gov. Summaries are generated through the Claude API under a fixed prompt template and a two-tier format that separates key points from detail, with rate limiting and retry logic for reliable API calls. API cost monitoring was built in from the beginning, and a reviewer feedback workflow is built into the platform for a later evaluation phase.

Solar production and utility billing in one picture

Solar Energy Intelligence

Live

A dashboard for a home solar installation in Maine. The solar company's app shows what the panels produce, and the utility bill shows what is owed, but neither shows both. This dashboard puts them together to answer what a homeowner actually wants to know: how many credits are banked, when they expire, and which rate plan would cost least. No AI is involved. The answers are calculated from Maine's billing rules and the household's real usage.

Design

A layout that shows a solar homeowner everything that matters at a glance, with the detail one level deeper for those who want it.

Build

A live dashboard that combines panel production with utility billing and updates itself every 15 minutes.

Teach

Visuals that show how solar credits build up and expire as it happens, and which CMP rate plan saves the most money.

Dashboard showing real-time energy production and consumption data

Real-Time Monitoring

Energy made and used, updated every 15 minutes, with daily, weekly, and monthly views.

Savings tracker showing net metering credit balance, cumulative savings, and ROI progress bar

Credit Tracking and ROI

Banked credits tracked month by month under Maine's 12-month expiration rule, alongside savings so far and payback progress.

Rate Plan Comparison

The same month's bill worked out under each of CMP's three rate plans, to show which one costs least.

Production and consumption data is collected every 15 minutes from the SolarEdge API and combined with utility billing records to model costs across three CMP rate structures. Credit tracking uses first-in, first-out queue logic to match Maine's 12-month rolling expiration under Net Energy Billing. The rate comparison shows whether Rate A, Electric Technology, or Time-of-Use pricing is most favorable for the site's actual production and consumption patterns.

Services

AI features with evaluation designed in

Summarization, extraction, drafting, and classification, with the limits and the acceptance tests agreed before launch.

Data cleanup and migration

Messy spreadsheets and exports turned into a clean, documented dataset or database your team can maintain.

Dashboards and analysis tools

Custom views of the numbers your team actually uses, with the calculations documented so you can see where each figure comes from.

Workshops and training

Practical sessions for staff and boards on what data and AI tools can and cannot do, taught by a former college professor.

Working with Katahdin Logic

Data and security

We maintain security practices appropriate to our scope of work and treat client data as confidential by default. What we build for you is yours: you own the code and the data.

How your data is handled, where it is stored, who has access, and what privacy or compliance concerns apply are all worked out during scoping and reflected in how the work is structured. We will tell you if we are not the right fit for a given compliance requirement.

Our use of AI tools

We use AI tools to write, debug, and document code faster. They do not replace our judgment about how a system is designed, secured, and launched. That is separate from whether the tool we build for you uses AI at all, which is decided with you.

Any questions or limits around AI and your data are settled during scoping. What is and is not acceptable is your decision, and we work within it.

Get In Touch

Does your organization have useful data that is a pain to get to?

Does your team lose hours to repetitive work that could run on its own?

Do you want to know what an AI task would cost at your scale before you commit to it?

Tell us what you are working with, and we will tell you whether we can help, and how. The first consultation is free.

Or connect with Allison on LinkedIn.