(2) user selects a custom date range -> we generate report for that

Move remaining hardcodings to .env. Here are the remaining hardcodings that claude pointed out (prod branch of 18th may 2026): https://drive.google.com/file/d/1_acrDmJKpCQmbfzKfWE-2SCeGNAtrsKW/view?usp=sharing

What we've maxed out in current build: * Voice quality – OpenAI Realtime (best in class) * Reliability – retries, timeouts, backpressure all solid * KB search gap – filled with natural filler speech * Language – English + Hindi auto-detect

Listen to recordings of conversation between pankaj sir and siddhant and further refine requirements; first listen to it yourself and then tell claude to do it: https://drive.google.com/file/d/1PjjN9cbeUb_-WZVHsEzd3qn9BoG1Tltf/view?usp=shar

Create a new API in node server that will be similar to /send_message but shouldnt have any: - preprocessing - postprocessing All it will be do take as input is a prompt and org_id: - search query generation - vector db search - call gemi

whatsapp, messenger, instagram, web-engage. blocker to: all tasks that require changes to the above integrations

add a retention period + compression to loguru

I found 10 ace tools that will completely rescue API bill. 1. Caveman Claude - Literally makes Claude talk like a caveman - Slashes 75% of output tokens with zero loss in accuracy Repo → http://github.com/juliusbrussee/caveman 2. RTK (Rus

Example: For Elefant, we have created resolution pathways for various cases. Both for pre-sale and post-sale. Outcome expectation is to allow CSMs (and eventually client) to automatically build that entire process themselves using Workflow

On the analytics page of the organization: - not all metrics will be relevant for each client - which metric will be relevant for which org? - add question doc for what all things should be there in grayed out sections We should ask claud

On dashboard's manage team page: add some way to configure emails too. Most of the CRUD required already exists in the backend. Mostly needs UI.

Suggested questions & counter questions right now are very basic. They need to drive engagement, they need to guide patients towards making the right decision, asking the right question, they need to help us find important details about the

Anytime a new org is created on prod; it should get auto-synced to test. Syncing limited to: - org data/config - vector data

"dev" branch is becoming a mess. "staging" branch isnt being touched at all. After a certain time, "dev" branch will diverge too much from the "prod" branch. There needs to be a release schedule where features from "dev" branches are merged

end_user agent NEVER just user

- ⁠PHASE 3: Once the user clicks the bubble -> chatbot needs to assess level of the user within upto 2 questions, and then in third question, ask user to select 1 of 3 goals - Learn, Practice, Test What follows needs to be the AI helping th

Depends on 301 to be completed first. Anytime a new bug is added to the bugs report sheet: - wait for 5 mins -> check if priority value is empty -> then use prioritisation rules to GET priority -> update the priority on the sheet Not sure

Get access from Shone Look at the notifications for 48 hours. Improve the setup to make it more relevant for you. Sync with Devanshu so he can track the bugs through Discord also.

- Number of members in team - Number of chats / AI msgs each month - Ticketing - Leads Qualification - Live chats support - AI messages - Channel integrations

(1) Hardcoded org check scattered throughout (Maintainability) The orgId === "ivfindia.com" check appears multiple times: Line 161, 290, 369 This could be extracted to a constant or config flag. We should move all such hardcodings to a comm

- [ ] add total AI messages, for all orgs - [ ] dropdown option, yesterday(default), this week, this month, last month

Example: https://integration.beyondchats.com/demo/aus-global-693fef89046e5/ This should show preview of https://ausglobaluk.com But since preview is not available, take a screenshot and show that

Check on your own, but then also discuss over a call with Shone so you're sure you're not missing anything

Option to update leads info, delete lead

- Just says failed with no reason why - Also, retry option hai but dikhta nhi hai.

Bot gave pound's unicode instead of actual symbol: https://beyond-chats.netlify.app/ivfmatters/103182?right_drawer_tab=answer_sources&message_id=529760

1. Now we only allow clients to train for TEXTUAL data. For remaining types: we notify admins like devanshu via email and show clients that the data has been added for processing and will take a few days, you will be notified on email once

See how background flask normalizes various links 1. The function will input an array of orgs -> get all their vector data -> normalize all read_more_links and inline links present in text 2. Generate a result csv containing new read_more_

-> Admins can add unlimited members to the org. -> They don't know what each role does. ->

After clicking "Open Mind-map" button, it should take the user to "DATA TRAINING STATUS" section and highlight the training data status that it points to.

- How can we make automated phone calls & answer them? - What Options available for Natural language audio conversation? - IVR replacement Check out: https://hoomanlabs.com/ https://vapi.ai/

Read last answer https://chatgpt.com/share/697d1816-494c-8002-94f7-f21bbc1ceae8

Node splits the final AI response and an array is returned from the /send_message API but in DB we store the AI responses as a string which loses the orignal formatting that node had intended for the user. This causes a sync issue between:

God level options for: - clear leads, chats, bulk messaging campaigns

Right now it shows, "data added for processing" but after these files get uploaded, it will take 2-3 days for processing because we have shifted to manual ingestion of these file types now. Uploaded the message to represent this. "Your fil

One user can login only on one device at any given time. Give popup option to logout from other devices

Start with ivfindia

Any message that is sent to the user should be helpful. The purpose of suppressing AI messages is to stop unhelpful messages like: "I cant help you with that" to be sent to the users but this is also suppressings helpful AI messages such

It is possible that some missing info was added in DB but currently no one updates its status. There needs to be a logic that would: - check if missing info still doesnt exist in db (simple keyword search) - if exists now: mark as irreleva

Right now we arent checking if some missing info doesnt exist in our entire DB. This can be as simple as querying for some keywords (not vector search) detected by the missing info analysis against the "text" column of the vector DB table.

When AI takes over conversations that were happening with human agent before, it messes up big time: https://beyond-chats.netlify.app/ivfindia.com/56784 The issue is: when sending messages to LLM providers like gemini and openai: we treat

Throw a validation error and present with options: 1. overwrite existing data? 2. show content thats already present + give option to Ignore

Currently we use gemini 2.0 flash for answering, if we were to switch models in the future then how we will not know how well the new model is performing because we arent analyzing response quality. The only analysis we do is when someone

1. Use existing missing info data to check: if it really doesnt exist in DB. [ONLY IF experiment 1 is majorly yes]:

Phase 1: 1. Match the `senderCurrPage` URL string against the vector's read_more_link (exact or prefix match). 2. Inject matching vectors in answer prompt (decide where to inject) Caveats: `senderCurrPage` may be null/empty on WhatsApp/In

Example: della resorts launches a new offer -> our AI finds which leads would be most relevant for this offer -> nudges them

We are not covering a bunch of cases: Case 1 moved to v2 Case 2: if an agent has already responded to the user then AI takes a backseat for 15 mins Case 3: if agent (or any admin) has the users chat opened on the dashboard then AI doesnt

Original ETA: 14 Apr

1. if remark doesnt contain clear instruction of what the GOOD AI response should have been then we need to ask the client. Contact devanshu or pankaj sir for it.

1. how to create MR (rules) 2. push to ci-templates's docs folder

User journey: 1. someone adds a new row on bugs report. 2. our system detects this new row. Parses "Chat Link" column for message_id and chat_id. 3. calls laravel API to fetch logs for the day message_id && chat_id happened. Add a param ca

for della resorts we dont respond to non-english queries. why we didnt trigger AI response generation shouldnt be considered AI suppression. In the SS, AI generation wasnt triggered because users message was non-english. We already store n

Original ETA: 27/Mar -> Contact Siddhant for KT of existing progress -> Already implemented to some extend in Node / Flask projects -> Understand the existing implementation -> Check if it works well for drmalpani.com, ivfmatters.co.uk, apn

(1) how to generate report of a month: not obvious enough

Edge case: No bridge exists for cross-platform chats. User comes on instagram -> clicks link -> goes to website -> talks to website chatbot

Listen to this recording first to understand what a landing page is and its purpose: https://drive.google.com/file/d/1Kc0R8OuLfcCdf94EoiNOBoD-MjtOaMWj/view?usp=sharing 1. create brand guidelines document 2. search what a landing page is +

For example: If someone from elefant directly DMs on messenger the do not let the AI reply for 15 mins. 1. try to reproduce it 2. debug why we are not tracking: are we not receiving Scenarios: case 1: staff messages a new customer (we h

Requirements unclear; we already mark pending tickets (stale) as resolved every month.

Sub task 1: document all of the external integrations Sub task 2: start with ones that our clients actually use Is whatsapp integration working fine? Is insta? Is website? Is live chat? Also if AI responses is working (Phase 2 planning to

store 3 options that the user gets + show them on dashboard

Remove all OpenAI voice integration Only have HeyGen-led video +video for real-time sync No compromise on lip-sync + audio sync - Still linked to BeyondChats send-receive msg pipeline

Investigate and fix Original ETA: 13/Apr - First verify if this is still true because siddhant had put "build" folder in .gitignore. Earlier we were committing the build folder too which caused the .git folder to be so large in size

-> Pre-created templates - AI decides when to send which template to which user

Token expires in 1 hour, after which the client will have to re-open the whatsapp dashboard from the admin dashboard again.

When we add new members to orgs, we have option to add their name. So basically anyone can change name of any user by adding them to org.

Phase 1: Let's track in DB. Phase 2: Display in dashboard w option to enable / disable on per org basis

Video looks good. Some suggestions: 1) Send email notification to ticket creator on replies / resolutions 2) If resolution not accepted, send another reminder 3 days later, then auto-mark as resolved after 7 days. 3) Tickets have a soft dea

Original ETA: 02/Apr Chat Embed Desing.html ---x--- Niraj will share code + live link https://sensational-stroopwafel-6ec1c8.netlify.app/ Pankaj will get it verified by Simran for any final improvement needs / reduction Pavan will implemen

CI pipelines of all projects will have to be updated

(1) Can use beyondexams whatsapp number for this. (refer to MetaWhatsapp::sendMessage) FYI: we already notify on discord for all prod based merges

All of the projects need to be moved to pankaj sir's namespace - ml-node, admin dashboard, embed are still on shone sir's namespace ml-node: https://gitlab.com/Shoney13/beyondchats-node dashboard: https://gitlab.com/Shoney13/beyondchats-n

Task Details: Many patients message ""call me back"" without context. Need to design a structured bot response that captures: name, patient type (new/existing), nature of query, preferred time to call. Coordinate with Pankaj/Siddhant on the

- remove lint (consumes alot of minutes and can be done locally) Any other optimizations?

- if bot can correctly answer from the sheet with the info that the user exactly asked for such as "where is my order" then: --> do create a ticket -> assign it to bot (laravel config has bot's user id) -> mark it as resolved -> also set r

Is it possible to obtain costs/our remaining creds via an API if these providers expose them? otherwise we may need to store versioned prices in our DB ourselves We dont want to give dashboard access to each service for obtaining costs/ava

Prompts available in node server and also in database On dashboard, check dev options -> Edit org details - some prompts (org_description, personality prompt, unsure msg, closure_msg) can be changed there End goal expectation is: We need a

Make it prod ready ---x--- Implement the following for ivfindia: User says: "I want to buy xyz" -> we show a razorpay embed button -> user clicks -> pays If payment success -> chatbot acknowledges that in some way (should be a different a

Project: Embryo Image Analysis for predicting embryo quality / development stage / implantation potential. Do some research. There are research papers that share code and some that share the model. Test out the codes locally and check whic

Check Elefant daily emailer + weekly analysis report And create new weekly emailer

Ensure all features available in desktop view are available in mobile view. We don't want to have a "responsive" web app as the dashboard. Assume we are building an android app. Mobile view of the dashboard web app should be a completely se

https://beyond-chats.netlify.app/medicalappraisals.org.uk/112887?right_drawer_tab=answer_sources&message_id=636032 Recently a system_note was added in /send_message pipeline whose purpose is to discourage bad behavior in sub-sequent AI res

https://docs.google.com/spreadsheets/d/1F4y7g0n198LWmZq5oI9oWIG5w5erW4nJB56lAqiavOk/edit?gid=763382465#gid=763382465

(1) First verify whether our bot is still repeating its answers: - https://medicalappraisals.org.uk/ - https://ivfmatters.co.uk/ - https://ivfindia.com/ we received feedback from the client that when the same question is asked multiple ti

Dynamic welcome message + welcome video clash. Switching page immediately shows multiple dynamic msgs. - 30 secs gap is not being followed. Check DrMalpani When I reply to dynamic welcome msg directly, there is a component visible behind

- explore what it is: https://ai.google.dev/gemini-api/docs/caching and https://developers.openai.com/api/docs/guides/prompt-caching This will require prompt re-writing where we may have to move all of the static context at the start. IM

Check app script in this sheet MediXcel -> Dr. Malpani Also shared API docs over email We want ALL patient data - files, doctor notes, treatment details, etc. in the sheet itself. Add button in sheet to re-sync latest medixcel data

Input = patient email (thread) + files if any Output = Doctor's reply + future thread thread: is a conversation ---x--- Ram has built the email syncing repo Check Task Updates of Row 115 1. coordinate with ram to get the dataset + prep d

Sent / received Thread-id mail-position-in-thread category of mail sent-by to / cc / bcc Date sent body subject Files present Download links to files

Too many org-specific hardcodings across the codebase. Phase 1: We need to control all this through env files or config files Phase 2: Manage all Role-based and Plan-based access controls directly through Laravel APIs and DB-storage

https://gitlab.com/sid6i7/beyondchats-node/-/blob/main/scripts/ai-review.py?ref_type=heads We use this script to automatically review MRs but it has several issues: - AI often gives positive feedback instead of flaws - AI doesnt consider c

in searchVectorDB.js, there is no filtering out of vectors that have 0 as their cosine similarity score.

Currently we use single LEAD_SCORE logic for all orgs which isnt right. Lets create custom LEAD_SCORE logic starting with elefant and della resorts. 1. Analyze chats and leads dashboard for elefant and della resorts. Think in terms of esca

Currently enabled for Cordelia Cruises

-> Talk to Claude + chatGPT. Show it existing analytics options - even the disabled ones. Also share all features we have (The original executive summary / claude.md mention these) -> Create a documentation of which stats needs to be added

Demo: https://drive.google.com/file/d/1e4XTceRee1v9SuiQoPWgabykWwPqO_s8/view

- use their sitemap to detect which pages were added/updated in 2026 + dec of 2025 - scrape those pages again using the scraper and config already defined - perform data cleaning if required

Show remarks, view chat option, option to add to conversation, posted on, posted by, resolved on, resolved by, etc. Check how Google or Microsoft do it. We need a simplified version of that. Check here: https://support.google.com/mail/thre

As PWA apps - Needs to work in both iPhone as well on android IIRC, Dashboard app was not working well - loading chats, joining live chat, notificaiton, etc.

User might join from multiple devices / channels. But DB stores only one channel per user.

https://chatgpt.com/share/697352e9-06c8-8007-8c7c-4eecd3bed086

Change is too big. We should make sure it is rolled out slowly. Rolling should be done at org-level. Some work is already done: https://gitlab.com/sid6i7/beyondchats-node/-/merge_requests/20 - whats solid about the previous work: intents a

1.⁠ ⁠Chatbot will send timely reminders to people whose egg freezing payment is pending. 2.⁠ ⁠⁠Chatbot will send timely reminders to people who have booked a consultation call. 3.⁠ ⁠⁠Chatbot will nudge patients who are getting treatments at

Build a frontend dashboard which should have the following feature: - A page to import leads which should expect a CSV to import a set of people/leads into the platform and save it for future works it should handle all the redundant cases

Chat-Embed project "How is your experience so far?" top bar needs to removed

Elefant wants ki agar generic queries puche user - Will you be my girlfriend? Do you like pizza? who is the PM of India, etc - ekdum befizuul questions - to chatbot thoda cool responses de - so that user ka experience bhi interesting rahe,

visit ivfmatters dashboard -> remarks -> unresolved and document: 1.⁠ ⁠which have been resolved and are no longer an issue 2.⁠ ⁠⁠which havent been resolved: propose solutions for this Note: try to reproduce the “unresolved” remarks. If yo

visit ivfindia dashboard -> remarks -> unresolved and document: 1.⁠ ⁠which have been resolved and are no longer an issue 2.⁠ ⁠⁠which havent been resolved: propose solutions for this Note: try to reproduce the “unresolved” remarks. If you

For IVF India, on their website, the costs are already mentioned in dollars as well as in INR so for people in India costs need to be shown in INR but for rest of the world cost should be shown in USD. detection if there is ANY price in ai

Chatbot improvement ideas: https://docs.google.com/document/d/1u6OMmkAhx-AtzbHtPMkb4dxTmdCF2YpJrOgwe_3ZcsU/edit?usp=drivesdk Frontend note: First we will need the students to take notice of the chatbot: 1. Create a new API timilar to /gen

Claude Analysis of the problem: pusher-connection-analysis.md Revisit: is our usage of pusher across laravel, embed and dashboard not properly optimized?

Original ETA: 25 Feb (Delayed due to other real-time tasks taking priority) Patients can upload their case documents, (DONE) Chatbot analyses the reports, and shares insights, (DONE) And allows for context-aware chat Also saves the file as

Discuss with: JayaShankar for leads dashboard Siddhant for what's already done

in a more reliable / maintainable manner (not just copy-paste to env)

Check beyondchats mail

Original ETA: 10/Feb

Adding remarks should allow for newlines and longer texts Original ETA: 7 Feb

https://gitlab.com/Shoney13/beyondchats-embed/-/commit/0e321ac62c4177a553062ae410fbc08b462f08ef/pipelines?ref=staging

Lots of irrelevant logs. Logs don't have correct API URL. Internal crashes need to be detected, logged and tracked MUCH better etc -> Bailleys crash on Della went undetected Steps: 1) debug highly recurring errors 2) check if some errors

End goal expectation: Nazeer knows ins and outs of all the features that we offer

1. find which projects + branches out of dev/staging/prod dont have a CI/CD 2. add CI/CD voice-service: prod

Devanshu's Sheet should get auto-synced