Extract relevant information from complex database structure (including many collections and arrays)

How To
, , ,
1

Hey there,
I try to build a virtual assistant that can help me do my things. :wink:
One function of it is, that it includes user specific knowledge from a database into the process of understanding user messages (and derive the right actions from it).

This is the structure of the database (already shortened and flatted):

[    {
        "PersönlicheInformationen": {
            "VollstÀndiger Name": "",
            "Geburtsdatum": "",
            "Adresse": "",
            "Kontaktdaten": [],
            "Beruf": "",
            "Ausbildungen": [],
            "Kritische Lebensereignisse": {}
        },
        "Beziehungen": {
            "Verwandte": [
                {
                    "VollstÀndiger Name": "",
                    "Verwandtschaftsart": "",
                    "Geburtsdatum": "",
                    "Adresse": "",
                    "Wichtige Informationen": []
                },
                ...
            ],
            "Partnerschaften": [
                {
                    "VollstÀndiger Name": "",
                    "Geburtsdatum": "",
                    "Wichtige Informationen": [],
                    "Aktuelle Partnerschaft": true/false
                }
            ],
            "Freunde": [],
            "GeschÀftskontakte": []
        },
        "GesundheitUndWohlbefinden": {
            "Offizielle Diagnosen": [],
            "Gewohnheiten": [],
            "Schlafmuster": [],
            "Lebensstilziele": [],
            "AktivitÀten": []
        },
        "Persönlichkeit": {
            "StÀrken": [],
            "Herausforderungen": [],
            "PrÀferenzen": [],
            "Persönlichkeitstests": [
                {
                    "Datum": "",
                    "Bezeichnung": "",
                    "Testergebnis": "",
                    "Bedeutung fĂŒr den persönlichen Assistenten": "",
                    "Ergebnisdetails": ""
                },
                ...
            ]
        },
        "Interessen": {
            "Beruflich": [],
            "Privat": [],
            "Lieblingsserien": [],
            "Sport": [],
            "Reisen": []
        },
        "AktuelleProjekte (privat und beruflich)": [],
        "Ziele (privat und beruflich)": [],
        "Herausforderungen (privat und beruflich)": [],
        "Besondere Verhaltensweisen": [],
        "AssistentenPrÀferenzen": []
    }
]

I now wanted to use date of this for a GPT module to derive with relevant information for another (later coming) GPT module, so that that one has all relevant data to do its job. Concretely, that other GPT module shall derive the user intention to send this message (as a first step of suppoting the user on getting there). So something like (in the user message): “Pleaes analyse this data and create a JSON object that includes all information that might be relevant for understanding the intention of this user to send a specific message. Please use JSON blabla
”.

Unfortunately, I have some issues here:

  1. The deeply nested collections can’t get analyzed specifically and are interpreted as “Collection” only, but not the content. Also toString didn’t work so far. Can you help me understanding, how I can fix this, so that the data gets accessible? I guess, the structure could be fully destroyed, since GPT is pretty good in still getting the data out?
  2. When I try to play and manually enter data, the GPT module only focusses on the obvious, but doesn’t get into deeper analysis, e.g. how personality traits/ personality testing results, motivators, strengths, etc. might play a role here.

Thanks, can’t wait to get support here (after two days of playing around and not getting any further). :slight_smile: