01
Assume the data is wrong
Names get misspelled. Dates get rounded. Instead of demanding an exact match, the software asks how close two records are and scores it.
I built forensic software to see the patterns humans haven't connected, finding links across thousands of records that no person has the time to compare. It spots when two records (missing person vs unidentified bodies) are describing the same person - even when the name is spelled wrong, the dates don't line up, there are discrepancies / inaccuracies, or the records live in separate systems that never talk to each other. It reads geospatial data too, such as how far unidentified remains were found from an Indigenous reservation, and it looks at a mass of data points across databases all at once, weighting the probability of every connection. My software finds it. I quickly realized that the software analysis engine I built can be used on databases across many industries to see patterns and irregularities. This opened up possibilities in government, banking, construction, small & medium businesses- the list is endless.
It started with a cold case in California which took 40 years to solve. Turns out, two records sat unmatched in a database due to a misspelling in the person's name. That showed me the real problem was the databases, not the evidence. I built the software and tested it against Florida missing person and unidentified body records. The same problem shows up in almost every business database, regardless of industry. This software engine looks at patterns in a unique way to quickly identify errors.
Someone types a name wrong. A date gets entered as a placeholder. One record goes into one system, the other goes into another. You only have partial or inaccurate data. Now you have two files describing the same person, and nothing in either system will ever connect them.
Most software only compares records when they're first created. Nobody goes back and re-checks the old ones against everything added since. So the match just sits there.
Same technique, lower stakes, cleaner data. The commercial version is where this research is headed — it isn't there yet.
Names get misspelled. Dates get rounded. Instead of demanding an exact match, the software asks how close two records are and scores it.
The software doesn't check records one pair at a time. It takes in a mass of data points across separate databases all at once and finds the connections no one has compared.
Geospatial data counts. The software measures distances that matter, such as how far unidentified remains were found from a tribal reservation, a key factor in cases involving missing Indigenous people.
Name, dates, location and description each carry a weighted probability, combined into one score. A weak name plus a strong location can still be a real match, and every score comes with the reasons behind it.
Two records in California, filed the same month, 45 miles apart, one letter different in the name. The case went cold for 40 years and was solved this year. Both records had been sitting in the database for the last ten years. Nothing in either system had ever compared them.
That was the point. The evidence was already there. The databases just weren't talking. I built the software to close that gap, then tested it on real Florida records. Names below are made up. The case is real.
The thinking behind this research drives every Innis engagement: assume the data is messy, weigh all of it, and show the work. Every service has a fixed, published price.