There isn’t a reliable fully automatic solution for the level of specificity you want. Generic image recognition may identify broad classes, but not consistently enough for exact subjects like a particular building or address unless you invest heavily in training and checking results.

The most practical workflow from the answers is to add structured notes while shooting, then match them to photos by time:

• Use a note-taking app on your phone with timestamps.
• Record each new subject/location as a short note; voice-to-text may help where practical.
• Match each note to the corresponding photos by capture time.

For newer images, GPS can also help:

• Extract EXIF latitude/longitude with a tool such as exiftool.
• Use reverse geocoding or location-query tools to suggest what is at or near that position.
• Treat this as an aid, not a final answer, since camera position may not equal subject position.

Your current in-camera markers (photo of a map, hand shot between subjects) are also sensible as manual separators. In short: combine timestamped notes, GPS lookup where available, and manual review. That is likely to be faster and more dependable than relying on image recognition alone.