ML/AI/LLM

10 mins

Yes I will

my invite expired

@54125194641609 can you invite me to email sync repo again?

Ek aur tool bheja tha Emon ne 10x wale group par shyd, voh better hai ig

Graphify bulky ho jata kaafi

oh okay. Seems helpful for planning stage only

everytime you add somethinng to codebase

graphify --update

is the KB graph updated every time a code change happens

how does it handle code updates?

In our codebase it can look up the codebase easily without waiting for those Greps and writing those commands to find something

Helps a lot

Graphify

@54125194641609 did you try any of these?

We can use this, it reduces ai agent's token useage by Turning codebase into a graph and then uses that graph, it reduces tokens to grep and read the grepped contents and makes it faster to find relevant files

https://github.com/safishamsi/graphify

we can use this to reduce token consumption on our ai coding agents

https://github.com/rtk-ai/rtk

The best medical care chatbot should be able to respond any medical question. Test it for that please

@102095734710467 can you test this chatbot ?

video bot: https://github.com/sid6i7/VideoChatbot

Emon69420

@54125194641609 whats your github username?

@102095734710467 thebestmedicalcare's chatbot can be tested here: https://integration.beyondchats.com/demo/the-best-medical-care-69724626d3a66

It’s a research task would need to connect before moving

why didnt you move to the next task?

Yes

yes i did, were you waiting for me to review it?

Did you receive the mail?

@54125194641609 updates?

-> Rather than not replying we will steer the conversation and deflect -> Rather than skipping negetive we will reply if related to dr priyanka -> Use a no. of message based regex relevance gate -> Update prompt from prod

🖕 | 🍆 | 🍑 | 💦 | 👅 | 🍌

.

Actionables for dr Priyanka

https://app.notion.com/p/actionables-392a4e78beb98072a9ebdf6b505451d7

Only suppression will take place if ai is giving own beliefs Share actionables be it for dms or be it for comments Make prompt injection better it’s too long right now for example micro cases

..

Figuring out how we can link the email’s add missing data to missing data so that we know missing data is added

@54125194641609 what are you working on today?

okay

2nd last is creating one constant that will be used at other places

The commits before that is are all model bumping

Last commit in the mr is for library update

is the migration and mode bumping in the same MR?

what should be the merge order of MRs?

Yes checked

are all APIs still functional?

@111561876209776 I updated the Gemini library on bg flask

Addressed

dr priyanka node

there is a minor fix i pushed on this branch

any short query in the language other than hindi english and hinglish will get rejected

long queries will work fine

so any short query with symproms will get rejected and marked as irrelevant <4 words

https://docs.google.com/spreadsheets/d/1FLtUTz2ERZl1_B-yucCYG7Ud5p-ezccSbgKtoaXKnKg/edit?usp=sharing

Okay let me try

Updated the MR

also, can u run a few more tests where: - TEST 1: input is noise/irrelevant chit-chat -> relevance checker fails (and returns TRUE) I want to see how our comment replier handles false positives where the relevant checker fails - TEST 2: input is relevant -> relevance checker fails (and returns FALSE) I want to see how our lead scoring handles false positives where the relevant checker fails (we can then include this in the daily mail for dr priyanka and these HIGH intent leads are uncontacted.)

not cool yr

Mr

I’ll update my

Umm wait

in your MR

we have switched the model to 2.5 flash for every single org

It’s a non contradicting prompt should work with both flash and flashlite

I thought it was flashlite

Huh

didnt we decide that it should be 2.5 flash

in custom lead scoring MR for dr priyanka, why have we changed the model to 2.5 flash lite? https://gitlab.com/Pankaj-Baranwal/background-flask/-/merge_requests/369/diffs

Where do we store elefant wagera ka ?

where do we store it?

Like customer support

Itent classification

do we store intent_category anywhere? we seem to be extracting this from the leads prompt for dr priyanka

@54125194641609 requested changes in model bumping to 2.5 flash in bgflask: https://gitlab.com/Pankaj-Baranwal/background-flask/-/merge_requests/368

okay thanks

Added

Adding it to

I shared it after all the sheets separately sorry

Wait a second

Huh

its not in the task updates

where is the sheet?

Yes we did that

for dr priyanka

For specific org

I think we did

@54125194641609 did we not test custom lead scoring?

alright

thank you. will evaluate it before 4 today

updated the sheet wiith result

Updating

I will do a final review by 4 once emon shares the sheets. Will have it live anyhow by 7, this time I want to roll it out to respond to only a sub-set of comments/dms so we can evaluate production properly

@111561876209776 ETA for DrPriyanka final update?

where can I review the final responses for dm replies/comment replies/relevance checker

@54125194641609 dr priyanka customization task row isnt updated with the new result sheets

also brainstorm what else can we suggest other than just blogs

we do not do any action on missing data end goal -> missing info is captured somewhere weekly mail is for some organisations, we tell about missing info to the client in weekly mail but we do not suggest anything to the mail, update the content in the mail we extract categories/ services we can suggest user to take actions on basis of this Laravel missing info to mark info as added, design the system for this but How do we check if some missing info is now added??? minimal llm calls for this

@111561876209776 link ?

Yes

can we connect by 9:30?

@111561876209776 when can we connect? Start working on actionables for missing data pts , i want to connect regarding this

In email-sync directory

Under export/drmalpani@gmail.com

Download all three files, these files will be saved on the Dev server

Download from frontend

And attachments.csv and dataset.csv is to map what message had what attachment

And how will I get all the emails on my local device for filtering out pvt emails?

A different folder will be used to save attachments

The emails will not be saved a whole json file will be saved with emails

Yes it will

And I hope it will fetch attachments, entire threads, all cc, bcc info, sent mails, etc

But where will all his emails be saved?

Thank you

@54125194641609 dr priyanka's org created on dev: https://dev.beyondchats.com/dashboard/dr-priyanka-maurya-homeopathy-6a422e4b3e68b

@111561876209776 kar Dena ye

and there was 1 flawed in the 150 marked with red

and intents

- "medical" — the user describes a health condition / symptom, asks whether homeopathy helps a condition, or seeks treatment / booking for a health reason. - "clinic_information" — clinic logistics & service discovery only: address, timings, fees, online / phone consultation, medicine availability, treatment process (no medical symptom stated). - "social_work" — non-medical help: harassment / abuse / safety, family disputes, vague distress cries, civic / locality issues, NGO / social / women's-commission grievances. - "noise" — greetings, political, religious, festival, emoji-only, r…

rubric that was followed

HIGH — the user shows real, established conversion intent. ANY of: - States an ACTUAL health condition, symptom, or disease that they (or a specific person they are asking for) HAVE. Example signals: "ghutne mein dard hai", "thyroid hai", "kamar dard", "bhukh nahi lagti". - Asks to BOOK an appointment, asks to be CONTACTED / called back, shares a phone number for callback, or commits to starting treatment. - Asks to MEET the doctor FOR A HEALTH REASON (a health condition or treatment is present in the chat). - Harassment, sexual / physical / domestic abuse, family disputes, o…

https://docs.google.com/spreadsheets/d/1Jleq0FZxC6rouoow3Nk1QEKltFagWtkr0B-_eBPgVFU/edit?usp=sharing leads

it would create a dump for the org

transfer function MR made, https://gitlab.com/Pankaj-Baranwal/chatgpt3_backend/-/merge_requests/1198

@102095734710467 please use this link to fetch emails for dr malpani, also after logging in you will be redirected to the default page of dev.beyondchats.com but if you go back to this link you will be logged in

https://dev.beyondchats.com/dataset-builder/

Okay it’s working

Okay

added

https://dev.beyondchats.com/api/dataset/oauth2callback add this

@111561876209776 remove local host from authorised uris

alright

claude is back?

tell claude on dev to setup reverse proxy on it

as far as i understood the whole server is path routed

@111561876209776

http://localhost:5859 this is the link for the local host in the dev

how do i see this on domain?

and the packages required

i have cloned and setup env vars

<media>

my side of the webapp logs into the gmail gets the dataset and then use it to finetune

added

creation of dataset is what we are doing afterwards. oauth is needed for gmail integration

we just need this to create a dataset

/dataset/oauth2callback is a very bad path name, doesnt explain what it does at all

han iske bina local host bhi kaam nahi kar raha

localhost?

@111561876209776

Google Cloud Console → Credentials → your OAuth client → Authorized redirect URIs.

http://localhost:5859/api/dataset/oauth2callback can you please add this to authorised uris

gate test with the political keywords

https://docs.google.com/spreadsheets/d/12jaQ_AJvfOuTukwJ5L2PXEhv0Nbt_4Zqc7PVyLQ5pzA/edit?usp=sharing

for longer queries

https://docs.google.com/spreadsheets/d/1zE_ueHyzsmvfcY2oSTlwoJ-peOBIp6a15_htJk2FoIs/edit?usp=sharing

<media>

can someone confirm indian flag comes in the output i guess windows does not support indian flag

after updates for comments

https://docs.google.com/spreadsheets/d/1pd9nRos2xjvvAKMUIo3bbW2KkHJJGCE8LqaJC6tzV6M/edit?usp=sharing

please add org269 data to the dev redis

@111561876209776

skip when under 4 words then skip else pass through gate, emojis are not words and non alphanumeric are not words skipped waalo mein emoji reply aaenge only if words, and like the comment we are growing engagement hence we need to engage back Mujhe aapse baat karni hai and similar are relevant prasing ka reply thank you jaega Female gender wise replies jaenge lead analysis for ig chat, user shared link do not process we should also give importence to her other charecterstics such as influencer and social worker key info only on user data not ai messages

Please let me in

Joining in 1 min

I’m in the meet

having lunch 🥲, ill join in some time

https://meet.google.com/jeb-qkni-tcf

Alright

Ok. Can you schedule a GMeet for 01:30 pm?

I’m available from 1

Batana when you can be available for a call to discuss this. @111561876209776 you should also join so you can review next version

Everything.

Overall or the replies?

@all I saw that dev server was using opus model. PLEASE switch to sonnet model for most usecases!!

we shouldn’t have github / gitlab logged in on any server. Hr baar manually creds daalne chahiye

Very poor quality

also @102095734710467 our claude limits hit 100%

do we have github logged in on dev server? also do we have a credentials.json in dev server in any repo for gcp?

@111561876209776 for Dr Priyanka

https://drive.google.com/file/d/1nRNd66b5BNHb1TtlX2T2wbG7YWJyy0xy/view?usp=sharing comment replies sheet https://drive.google.com/file/d/1Zc2cOC-tuJJJUI4k9bdy4R4VbUX1WQ2u/view?usp=sharing Comment and replies relevancy sheet https://drive.google.com/file/d/18DUhrvJ7CvCW-f7gu-3QGDRnwi_LSetz/view?usp=sharing Lead score test result https://gitlab.com/Shoney13/beyondchats-node/-/merge_requests/340 Node MR https://gitlab.com/Pankaj-Baranwal/background-flask/-/merge_requests/369 Flask MR

Okay

@54125194641609 added task to row 512 for dr priyanka

for elefant, if someone asked a question on comments then we used to tell them to DM us. should the requirement be the same for dr priyanka?

Alright

Possible to get this ready by Friday?

This is high prio 1

Okay

Please make it easy for me to take action

Aim is for me to import the mail dump locally on my system. I will then check the mails, and remove all confidential info, then forward to you

So I can use it for getting the mail dump

And add dr malpani’s email as test user

Please set it up on dev server

On dev server

But don’t we already have all that set up?

Google cloud waala

You can do it on your device just to setup Google o auth

https://github.com/Emon69420/email-sync/blob/main/docs/SETUP_AND_RUN.md

Nah

So I’ll need to set it up on some server?