How we can work together
A focused first project.
A clear deliverable.
Investigate a policy or program, evaluate a technology, or build an AI and data workflow. We agree the question, deliverable, and standard of evidence before starting.
Before you build / AI feasibility assessment
Is AI the right fit for your workflow?
Before investing in a tool, assess the task, available data, and operational constraints. Compare AI with simpler approaches and identify whether a pilot is worth pursuing.
What you receive
A feasibility brief, a comparison of approaches, and an evaluation plan with criteria for proceeding, revising, or stopping.
Bring a workflow description, representative examples, and the decision you need to make. We agree the assessment scope and fee before work begins.
01 / Report-to-Dataset Pilot
Turn a report backlog into usable records.
Build and test a workflow that extracts agreed information, links it to source passages, and brings uncertain results to human review.
For research and operations teams repeatedly reading reports to fill spreadsheets.
What you receive
- A structured dataset with field definitions and source references
- A tested extraction and review workflow for the agreed document family
- An evaluation of errors, review effort, and processing cost
- What we need to start
- Representative reports, the fields you need, your current process, and a reviewer who knows the records.
- How we judge success
- Test on human-reviewed records held out from development. Compare field accuracy, important missed cases, review effort, and cost with the agreed baseline.
02 / Connected reporting
Make recurring reporting easier to trust.
Connect agreed sources, resolve definitions, and build a reporting workflow that makes missing data and inconsistent records visible.
For teams rebuilding the same reports from separate files or source systems.
What you receive
- A linked dataset and documented metric definitions
- A dashboard for the agreed operational questions
- Data-quality checks, a refresh procedure, and handover documentation
- What we need to start
- Representative source exports, the reports you use today, agreed measures, and an owner for each source.
- How we judge success
- Reconcile key totals to the sources, test missing and duplicate records, and verify the agreed reports and refresh procedure with your team.
03 / Research & policy evaluation
Give the next decision a stronger evidence base.
Study a justice program, public policy, or technology, and investigate stakeholder experiences with methods suited to your decision and the available evidence.
For public agencies, research teams, and organizations evaluating a policy, program, technology, or change in practice.
What you receive
- A study plan with agreed questions and measures
- An evidence review, survey findings, or outcome analysis
- A findings report, limitations, and a decision briefing
- What we need to start
- The decision you need to make, existing evidence and records, the people involved, and relevant deadlines.
- How we judge success
- Answer the agreed questions with traceable evidence. Explain what the study can establish, what remains uncertain, and the implications for your decision.
Research judgment + AI engineering
The tool and the question
belong together.
Research design, public-policy context, and technical delivery inform each other. We choose methods that fit the question, whether the result is a study, an evaluation, a dataset, or a working tool.
Explore our approachExplore the work / Connected reporting
See the data.
Understand the comparison.
Our team's work with FBI agency records brings source integration, completeness checks, and descriptive analysis into one interface.
Select a state and population band below. The results describe eligible agencies and show the limits of each comparison.
Data to decisions / Public-sector analysis
Agency crime trends by state
Select a state to compare agency crime rates.
Dashed states have fewer than five eligible agencies in the selected population band.
125agencies included
68excluded by data checks
The same agencies are included across 2022–2024 and both crime measures. Each year requires 12 monthly values and valid population data.
Illinois
2022–2024Offenses per 100,000 people served. Each agency has equal weight in the median. Results describe included agencies, not statewide crime rates.
Median rates, 2022–2024
Both series use the same scaleShading: middle 50% of comparison agency rates. This is observed variation, not a confidence interval.
Key findings
Among included agencies in Illinois, the 2024 median property crime rate was 19.6% below the median for eligible agencies in other states within the same population band. The median rate rose 0.1% from 2022.
Review reporting practices, service demands, and population changes when interpreting these results. The analysis does not establish causes or measure technology effects.
Methodology and data
125 of 193 city agencies in Illinois within the selected 2022 population band meet the data requirements. The comparison includes 2290 eligible agencies in other states within the same band.
Eligibility starts with city agencies whose recorded 2022 population falls within the selected band. Agencies outside that band or without a usable baseline population are not counted in the inclusion and exclusion totals. Data checks require full recorded population coverage and nonnegative annual offense totals. Comparisons are descriptive: agency type and population band do not account for all differences between agencies. Reporting changes and the complete-data requirement may affect results. Complete monthly records do not independently confirm reporting accuracy.
Rates use each agency’s annual population recorded by the FBI. The median summarizes agency rates with equal weight per agency. Percentage change compares the 2022 and 2024 medians; it is not the median of individual agency changes. The comparison excludes the selected state. Results require at least five eligible agencies in each comparison group.
| Year | Selected agencies: median | Selected agencies: middle 50% | Other-state agencies: median | Other-state agencies: middle 50% |
|---|---|---|---|---|
| 2022 | 1,066 | 683–1,842 | 1,551 | 889–2,471 |
| 2023 | 1,077 | 654–1,775 | 1,472 | 853–2,338 |
| 2024 | 1,067 | 645–1,792 | 1,327 | 763–2,141 |
Source: FBI Crime Data Explorer, agency roster linked to annual agency records. Source snapshot updated: July 15, 2026. Extracted September 26, 2026. Founder research and engineering work. Only aggregate results are displayed.
How we can work together
Bring a question.
See what we can deliver.
Explore how we can work together. We define each project with you, agreeing on the objectives, available data, deliverables, and measures of success before work begins.
| A question you bring us | What we do | What you receive |
|---|---|---|
| We have thousands of reports. Can we use them to investigate a problem?Discuss this project | Define the information needed, build and test extraction or classification tools, check errors, and analyze the resulting records. | A checked dataset with source references, an extraction and review workflow, and a report on accuracy, effort, and limitations.See the Report-to-Dataset Pilot |
| How do our results compare over time or with similar organizations?Discuss this project | Connect records, check reporting gaps, define appropriate comparison groups, and examine patterns and differences. | A comparison report and an interactive dashboard with documented definitions, data coverage, and limitations.Explore our FBI agency analysis |
| We are considering a technology. What evidence supports buying it?Discuss this project | Review relevant studies, examine how the technology was tested, and assess how those settings compare with yours. | An evidence review, a comparison of findings and limitations, and the questions your own trial needs to answer.Our review of AI-assisted policing Our team’s review informs how we distinguish promising applications from demonstrated outcomes and assess whether findings apply to your setting. |
| We introduced a program or technology. Did it improve results?Discuss this project | Study implementation and outcome records, using suitable comparisons where the data and study design allow them. | An evaluation explaining what changed, what can reasonably be attributed to the intervention, and what remains uncertain.Our camera-footage study Our team’s study identifies associations in complaint records. It illustrates why an evaluation must distinguish observed differences from causal effects. |
| What do employees, service users, or the public think about this change?Discuss this project | Design and analyze a survey about the specific change, including experiences, concerns, and differences between groups. | A survey instrument, a findings report, and recommendations for implementation and communication.Our research on public attitudes to AI Our team’s survey research informs questions about attitudes, knowledge, and institutional confidence. Its associations do not establish how to change public support. |
AREAS International
Working with organizations internationally
We welcome international inquiries for applied AI, technology evaluation, data analysis, and research collaboration.