Hi Nick, welcome back! Ryan was asking us if James and I received our FiveTran training and just wanted to follow up on it now that you're back
Hi Nick, here's the link to my work logs https://docs.google.com/document/d/1hZW7HIdr4r18vVUkqnpA0_QQwvHdQzNijLga5IFBtBE/edit?usp=sharing
please let me know if you cant access them
Ya having an issue on my phone
ok i gave permissions now
Just sent the first draft of the interview assessment, please let me know if i needs adjustments
(I still need to add parts concerning their Google Cloud credentials and GCP API Token)
I am getting Joe to give us a laptop so we can configure that before the candidate arrives
When you get in go get laptop from IT
examaccount@shield-legal.com / Friday5425@
Can you make sure itβs setup with vs code and ready to go?
Anthony told me theyβll deliver it to us within the hour
Once i get the laptop, I will copy the json credentials over to it
Once i get the laptop, I will copy the json credentials over to it
Hi Nick, so the calendar shows that we have App Dev Weekly scheduled tomorrow at 10:30
will we still need to meetup at 1:30 tomorrow?
Hey we have a meeting with Abe and Alan at 11
He is already kinda on one this morning and the deadline is probably shorter than I expected so he may have thought we were working over the weekend on it. So just prepare to really jump on this project. Ask as many questions as you can to understand all of the components.
Yea Iβm currently working on scraping the sample file still
240 Secondary Interview Completed or Secondary Interview Re-Triggered for ACTS-DL-Flatirons Chowchilla. These are the ones with documents uploaded to LawRuler
Abes project description: The litigation has determined that each plaintiff shall file a short form complaint. The format of the document is non-negotiable and must match to what the court ordered in the provided template. The expected values must also match to the expected format from the court provided form because this will be filed with the court. The data may have already been collected by the vss team during secondary interviews however the format of the data is probably incorrect for this document. This project will require data gathering from shield legal's intake and vss secondary interviews and writing scripts to transform the data into the correct fields or alert us to requiring the vss to follow up with a client to get missing data. This project will also require tracking of what has been done v what will be required. As an example of a script i am thinking of....writing an if statement to look at each cell to see if it will answer the question as a yes. Or writing a more in depth if statement to look at three different cells to see if the answer is a yes
I performed a quick test with ChatGPT using the account James gave me, and it gave me the expected answers to those checkbox questions (e.g. kissing, vaginal penetration, etc)
so I will go ahead and integrate LLM for that searching process
Devon told me that I would have to ask Abe for the "why" part of question 4.
So I'm thinking of requesting a blanket statement on that as well?
i spoke with Alison and she said that IF a case is Completed then they have all the info on it - including the yes/no answer to "Did you file a report?"
And we would need a blanket statement for "Why didn't you file a report?"
we would only want to look at the Completed ones any ways
Can you send me an example output of what you've done so far?
[
{
"Plaintiff_Name": "Andrea Marie Baca",
"Individual_Case_Number": "",
"Plaintiff_Action_Type": "new_case",
"Add_On_Petition_Filing_Date": "",
"Defendants": {
"State_And_CDCR": false,
"DOE_Number": "",
"Named_Individual_Defendants": [
{
"Name": "",
"Correctional_Officer": false
},
{
"Name": "Plaintiff does not recall name of abuser and could recognize in a photo line up.",
"Correctional_Officer": "correctional officer"
}
]
},
"Year_Of_Abuse": "2010",
"Location_Of_Abuse": {
"CHOWCHILLA (CCWF)": true
},
"Plaintiff_Incarceration_Status": null,
"Number_Of_Assaults": "4 incidences",
"Details_Of_Assaults": {
"Kissing": {
"Occurred": true,
"Body_Parts": "breast, mouth",
"Quote": "he stopped to talk to me and proceeded to feel my breast and kept feeling the rest of my body. Plaintiff reported the abuser touched her breast with his hands, as well as with his mouth."
},
"Fondling": {
"Occurred": true,
"Body_Parts": "breast, vagina",
"Over_Clothes": false,
"Under_Clothes": true,
"Quote": "Plaintiff reported the abuser touched her breast with his hands, as well as with his mouth and that he rubbed his penis against her as she was pushed up against a wall. Plaintiff reported the abuser grabbed her from behind and rubbed his penis against her as he touched her vagina with no gloves on. Plaintiff reported one incident of digital penetration with no gloves on."
},
"Vaginal_Penetration": {
"Occurred": true,
"Quote": "Plaintiff reported the abuser grabbed her from behind and rubbed his penis against her as he touched her vagina with no gloves on. Plaintiff reported one incident of digital penetration with no gloves on."
},
"Anal_Penetration": {
"Occurred": false,
"Quote": ""
},
"Oral_Sex": {
"Occurred": true,
"Quote": "Plaintiff recalled an incident of sexual abuse occurring outside in a blind spot when Plaintiff was sitting down and reading when the abuser approached her and forced her to perform oral copulation. Plaintiff reported the abuser grabbed her head and pulled her toward him while telling her \u201cyou know what to do with it\u201d."
},
"Masturbation": {
"Occurred": false,
"Quote": ""
},
"Other": {
"Occurred": false,
"Type_Of_Abuse": "",
"Quote": ""
},
"Additional_Description": ""
},
"Govt_Code_845_6_Medical_Description": "",
"Govt_Code_845_6_Why_Known_To_Staff": "",
"Search_Only_Explanation": "",
"Claims_Asserted": {
"Sexual_Assault": false,
"Sexual_Battery": false,
"Gender_Violence": false,
"Intentional_Infliction_Of_Emotional_Distress": false,
"Govt_Code_845_6": false,
"Bane_Act_Civil_Code_52_1": false
}
}
]
so ideally, there would be an array of records per case
the Claims_Asserted field just has placeholder values. I'm not sure Im going to keep it
Andrea Baca is actuallly such a case where the Plaintiff couldnt recall the name of the abuser
I asked Devon and Alison about No.9 and they said the answers to these are probably gonna come from Abe
Draft what your questions would be so I can forward to him please thanks
Joe was holdin me up so i found my own solution installing stuff
here's a sample of a filled out form ^ (it doesnt fill out number 9 filled out, and DOE fields)
my question for Abe is:
For Question 9 on the Short Form, are the answers to the checkboxes derived from the Plaintiffβs description of the incident? Or will they all have the same boxes checked?
```For Question 9 on the Short Form, there are checkboxes for claims:
Do the cases all have the same boxes checked? If not, how would we derive the answers to them?```
his answer was derived from answers
even with the Question 5, it doesnt extract at 100%, more like 85 - 90% reliability
An ai agent would have to be given the semantics for each of those clauses to be specialized for them
What do you need for the semantics?
What do you need for the semantics?
so for James's project, we fed it a long pdf doc about how drug tests are properly performed
THEN we gave it pdf's of the drug tests and asked the LLM to determine if there were things that the experiment missed or malpractice
but that could take a while to test since I would have to manually validate results per case
ok then I will tell him that: We are currently not confident in the LLMs ability to determine these. Even in #5 we are seeing 85-90% reliability. The model would need to be given semantic information to hopefully improve it. For example, in a different project we had to feed it a long pdf about how drug test are properly performed. Then we gave it pdf's of the drug tests and asked the LLM to determine if there were things that the experiment missed or malpractice but that could take a while to test since Josh would have to manually validate results per case. Do you have an input/recommendations for how to handle these or will we have to resort to some manpower to manually review?
We could suggest that given a document filled with the semantics as to what constitutes any of these acts (e.g. GENDER VIOLENCE), I could see how well it does
But I don't know if that's within the bounds of his timeline
We are currently not confident in the LLMs ability to determine these. Even in #5 we are seeing 85-90% reliability. The model would need to be given semantic information to hopefully improve it. For example, in a different project we had to feed it a long pdf about how drug test are properly performed. Then we gave it pdf's of the drug tests and asked the LLM to determine if there were things that the experiment missed or malpractice but that could take a while to test since Josh would have to manually validate results per case. If we could get a document filled with the semantics as to what constitutes any of these acts (e.g. GENDER VIOLENCE), we could see how well it does. Otherwise do you have any input/recommendations for how to handle these or will we have to resort to some manpower to manually review?
From Abe: we will see what you have on monday and make a plan and figure out what we can v what we cant pull out plus how the human review will have to work.
side note: I'm still running into some issues parsing the ACTs docs - they dont follow a consistent convention (e.g. some answers are on the same line some arent, some answers are multiple paragraphs, some keys to fields are different, etc.)
Hey Alan came over to tell me that you have been using the ACTS only version. We need to use the secondary interviews from DL - Flatirons and ACTS -DL - Flatirons. Unfortunately, he told me the formats are different. Can you get with them tomorrow and they can show were to get the right ones.
Can you go find and work with Ralecia first thing tomorrow morning. They will make sure you got what you need and align
yea im currently talking to Devon to get the right sample docs
Good morning Nick, i've got a fever and will be staying home today. i'll keep an eye out for any emails from Abe in case he needs anything urgent though
hi Nick, I'm still not feeling well, but hopefully I'll be able to come in tomorrow
Just try to watch out and respond as you can
Thanks and you too