A researcher has continued work on a conversational assistant that answers questions about crime statistics by querying national datasets, saying the tool is improving but still imperfect. The project, currently online at NIBRSAI.com, is intended for eventual inclusion in a newly launched portal, CrimeIndex.org.
Scope and sources
The developer says the assistant must handle an exceptionally large and varied corpus of data. Sources described include:
- More than a billion rows from NIBRS (National Incident‑Based Reporting System).
- Millions of rows of current and historical crime, clearance and staffing records.
- Hundreds of thousands of homicide reports.
- Tens of thousands of hate crime records.
- Historical SRS data reaching back to the 1930s, with clearances and staffing data extending to the late 1950s.
| Dataset | Approximate size / span |
|---|---|
| NIBRS | More than a billion rows |
| Homicide reports | Hundreds of thousands |
| Hate crime data | Tens of thousands |
| Historical SRS / staffing | Back to 1930s / late 1950s |
Technical and user challenges
The author acknowledges both technical and human limitations. On the technical side, assembling multiple sources with different structures — for example, the Supplementary Homicide Report versus current crime feeds — complicates query design and accurate retrieval.
On the user side, the developer found that translating natural language questions into reliable SQL queries proved difficult. They also reported that many real‑world user queries exposed gaps in how the assistant interprets intent and selects appropriate sources.
"The assistant is at NIBRSAI.com and it is designed to answer your crime data questions while highlighting the limitations of the available data and avoiding analytic AI slop."
Progress and aims
The project is described as ambitious but iteratively improving. The developer emphasises that the assistant is not yet perfect and that some answers remain buggy or incomplete, but said gathering hundreds of real‑world queries helped identify strengths and shortcomings. The stated aim is rapid, accurate responses drawn from the combined datasets in a matter of seconds and to add the assistant to the Crime Index platform when ready.
The work highlights the broader challenge of making complex criminal justice data accessible and reliable for journalists, researchers and the public while preserving clarity about provenance and limits.
