- first iteration - very early days, just cutting and pasting between llm and spreadsheets to determine time zones and extract key words
- core addition - scoring just on time zones and
- core decision - never let score fully decide
- second iteration - improved survey, improved algorithm
- core addition - collecting data just as important as acting on it
- core addition - taking previous matches into consideration
- third iteration - matching algorithms plus ai matcing
- core addition - multiple years of matches as well as rematching
- core addition - tried published algorithms
- core addition - refined tooling for scoring
Learnings -
- human in the middle - essential
- skills based levelling was so crucial - someone comes in at a G12 from an acquisition and is really more of a mid level based on their skills but doesn’t know that
- Scaffold support materials to get folks talking
Wish list for future
- Need to migrate to goals or future needs as the number of prevously matched folks increases
- Dynamic processing of any kind of matching data - determine fields, classify them, build the scoring algorithm