DrMalpani webmaster

9) Analyst has been sharing pending tasks. Need to update its status in tasklist tracker Delayed. Couldn’t get to this today.

8) BeyondChats chatbot bugs fix - Gemini models are overloaded leading to fallback to openai model which hallucinated a lot. Need to figure out better fallback mechanism. Team is already working on this. Tried latest gemini model - google’s models are overloaded across all models, tried gpt-4o-mini but responses were hallucinating a lot. Need another couple of days to figure this out reliably. Team has also evaluated ai-twin rag pipeline as part of this task. Haven’t received an update yet.

7) AI Twin embedding finalisation Delayed. Other tasks took a lot more time than expected. Kal ya Monday finalise krenge ab.

6) DrMalpani Ads review + changes Delayed. Kal discuss krenge w Devanshu

3) SEOForge only tracks wordpress websites content. I need to extend it to track DrMalpani as well so I can re-use features like Tl;dr generation This turned out to be a MAJOR overhaul - even with claude code. Sb kuch scratch se bana hai Clinic website pe across Laravel, react, etc. Wordpress has a very well defined pattern. I need 1-2 more days before this can be implemented well.

2) Working on Phone calls automation service https://docs.google.com/document/d/1CpyyPpuvs7kXezoMazyVm-ylXm9mSssN/edit Market research is complete. Very opaque pricing across all vendors for SIP trunking. Clear pricing is available if we want human-agent led virtual phone call management, but not if we want to use our agentic pipeline Next week iske aage continue krenge. Only 2 days are allotted for experimenting now that we are moving away from experimentatoin for 2 months.

1) Brainstorm how to track usage + ROI for existing experiments Discussed w Sid (Product Lead). We brainstormed with claude and amongst ourselves. There are several options and each feature will require thoda sa manual intervention / claude-led automation to figure out what ROI to expect + how to productise tracking it at a per-feature level. We will take up 5 experiments that are already live and implement this in the 15 - 30 July sprint. I need to share which 5 experiments and what ROIs by tomorrow (Sat). End goal is that we have clear SOPs of “how to do this & what mistakes to avoid” in t…

How many of these have been completed ?

Meaning?

And there needs to be a clear gap between the end of the summary and the start of the article

Please don't use tl-dr because this is a medical website and patients will not understand this word. Just use a quick summary

The 5 latest articles on the website now have tl;dr section

https://www.drmalpani.com/knowledge-center/articles/why-measuring-uterine-blood-flow-is-pointless-in-an-ivf-cycl

<media>

Minutes: -> Analyst only analyses first 5 docs. Let it assess all. We can review costs once usage increases. -> IVF analysis happened for one Pranjal. But when Dr clicked on the link, it took to the generic page. -> Pankaj mentioned patient will be redirected to their medical vault. But he will confirm once if that’s actually happening. -> We also need to allow dr to go to the same link and see the report.

ok

The summary at the top is more important because it tells people what the article is all about

@179035711746252 The FAQs at the end of article - isn't that already enough as a Tl;dr? Why do we need a summary at the top as well

Tasklist: (across me and the team) 1) Brainstorm how to track usage + ROI for existing experiments 2) Working on Phone calls automation service 3) SEOForge only tracks wordpress websites content. I need to extend it to track DrMalpani as well so I can re-use features like Tl;dr generation 4) Tl;dr generation for drmalpani 5) MiceLabs website changes 6) DrMalpani Ads review + changes 7) AI Twin embedding finalisation 8) BeyondChats chatbot bugs fix - Gemini models are overloaded leading to fallback to openai model which hallucinated a lot. Need to figure out better fallback mechanism. 9) Analys…

https://x.com/pankaj_cic/status/2077971785190891944 https://www.linkedin.com/feed/update/urn:li:activity:7483737511001948160/ Shared on X & Linkedin as well

StartupMentors chatbot review - unbiased, used it as an entrepreneur

<media>

Thanks!

Tomorrow

When are you planning to add the Quick Summary to all the existing articles @102095734710467

https://www.drmalpani.com/knowledge-center/articles/you-re-the-captain-of-your-infertility-journey-your-doctor-is-only-the-navigator

https://www.drmalpani.com/knowledge-center/articles/what-do-you-really-want-a-baby-or-a-baby-with-your-genes

https://www.drmalpani.com/knowledge-center/articles/if-your-ivf-doctor-tells-you-to-lose-weight-first-read-this

Urls ?

Also we had decided that after patients asked five questions , we would ask them for an email and a whatsapp so we could share the transcript with them, but this has still not been implemented . I get irritated when things fall between the cracks like this.

The scans with the regular chatbot are far stickier and I think we need to learn from this experience and apply to the AI twin

Please specify dates

Already being implemented. Bohot saare custom workflows hain jo BeyondChats chatbot pe implemented hain which make it so good at clinic conversations. So AI-twin RAG-engine ko integrate kr de rhe w these pre & post processing steps. ETA is coming Tuesday for testing it on AI-twin first, then if it works well, Thursday tk same pipeline will be used on text & voice chatbot also

How do we implement this?

One of the high priority items is to make sure that the chatbots are conversational because right now the AI twin just answers the question but there is no flow at all

Will check later

Would love to understand how relevant this feels for your clinic use-cases

@179035711746252 ClinicOS demo videos, finally ready! 🤞🏼🤞🏼

https://www.youtube.com/playlist?list=PLXZ-er3FA5yQ

Why are these errors occurring ?

This answer is wrong because when patients are doing IVF they do not need to do an HSG to check if their fallopian tubes are open or not. Can you please correct this

<media>

<media>

Please add this to the task list

Quick Summary ⭐⭐⭐⭐⭐ Key Takeaways In a Nutshell What You Need to Know The Bottom Line ⭐⭐⭐⭐⭐

Dr. Malpani's Bottom Line Dr. Malpani's Key Message What I Want You to Remember My Advice in One Minute If You Read Only One Thing... The Most Important Point

Call the two-line summary this:

@analyst please share updated and pending tasks

We should consider adding a short two-line summary at the top of each page on the website to pull users in

Makes sense. You should be able to do most of this in 2 weeks since the tools are already live. It's just a question of stitching them together now

@179035711746252

Please edit it. It's outdated

Please share the demo for the ClinicOS

Figuring out how best to let it function alongside netcore.

I still need to update the email to ask@drmalpani.com though

https://www.drmalpani.com/ask-dr-malpani Analyst email tool landing page

@179035711746252 https://www.drmalpani.com/book_an_appointment.html Should we delete this page?

https://www.drmalpani.com/fertility-roadmap/roadmap.html

Claude hi use krwaya tha, after you pushed for it. I’ll revamp it. Was already working on a plan

I would suggest you use Claude for the writing and make sure that it uses my voice based on the original articles which I have written

https://www.drmalpani.com/success-stories/lakshmi-and-magan-s-story-overcoming-infertility-blame-malpani-clinic —> what kids of stories are these? these are not testimonials. they’re badly written and they seem like lies

Lazy AI writing

https://www.drmalpani.com/ - this is not working in your favor

Thanks!

<media>

No email as yet 😕

*Dr Malpani Weekly Performance Report: Jul 7 - Jul 13* =================== *OVERVIEW* *Paid Traffic:* 13476 users (↑ from 9325) | 3s Avg Engagement (↓ from 4s) | Top Cities: Patna (1433), (not set) (1038), Lucknow (700), Ahmedabad (696) *Organic Traffic:* 2340 users (↑ from 1936) | 92s Avg Engagement (↑ from 89s) | Top Cities: (not set) (103), Mumbai (94), Bengaluru (86), London (65) *Insight:* Paid volume grew another 44.5% (9325 to 13476) off the same YouTube-driven pattern as last week: engagement dropped again (4s to 3s) and the city mix is still tier-2/3 (Patna, Lucknow, Ahmedabad), s…

You can now resume emailing the analyst

You must have received it - for Mrs. Neha

But kal around 12 baje aapko mail bhej diya tha

The email-analysis service is spread across 3 different servers because it uses BeyondChats’ pipeline for replying to emails. So had to change logic across all three servers. And usme time lg gya

The issue was that we are using gemini-flash-lite model for data understanding in the email analysis tool. Google’s servers are getting over-used these days and it led to: “503 service unavailable” error.

Thoda zyada late ho gayi thi because a lot of things had to be changed on the backend.

Hi Sir

Done?

Sorted?

Yes, exactly. please wait. Checking

<media>

Wait, it didn’t work

You must have received the reply now

Thanks ! Please let me know when done. Have a consultation with them

ADding a fix

Instead of just skipping moving it to conversations view

We obviously dont want your mails to end up in conversations section. Thoda galat code implement ho gya tha, we were skipping your mail completely

<media>

No reply as yet 😕

What happened to the email I sent to analyst@ivfindia.com ?

1) StartupMentors.in -> Chatbot will go live by eod today (14 July) 2) TBMC chatbot -> Already answers from general world info. 3) ⁠We have decided not to ingest books at the moment. Dumping many books to AI actually becomes problematic in the long run because chatbot isn’t able to gather *insights* from the various books when answering people’s queries. It starts randomly clubbing info from random books because they _feel_ related to user’ question. Better approach would be to let chatbot handle people’s queries on its own, and when it gets stuck, we can ingest some books specific to those t…

What can I help you with ?

Poor English

How do you track user satisfaction? Before and after?

How will you document improvement?

Already on it.

The problem with new technologies is that they keep on improving, which is why we need to keep on remaining on our toes but the good thing is that, based on our past experience, we can do a much better job much faster

And if you are building a new pipeline, you should use it for the old bot as well

Or maybe we should build everything from scratch anyway since BeyondChats pipeline - while reliable - is old and cluttered. Hmmm, I’ll brainstorm about this and get back to you

These are all the things we have already figured out at Beyondchats. I will discuss w team to find a way to integrate the new RAG system with our surrounding pipelines for handling such cases

The bot needs to stop asking the same questions repeatedly

<media>

Makes sense. I've also noticed that on the landing page. But didn't think much about it as it's just a landing page and I understand what the intent was

i can see the Ai twin talking on the landing page but can’t hear the audio, perhaps having the option to hear as well could be useful?

hi doc, overall this looks pretty refined- one thing that i could think of was having the entire process recorded, from starting the voice call to ending it with maybe a consenting patient’s actual audio or a placeholder fake patient, and having it played on the landing page itself? that way the user can see exactly how their call will go

Saurabh is working on the embeddable AI-twin version. Wednesday tk it should be ready for website integration alongside chatbot for A/B testing - abhi issues a rhe hain when launched as an iframe.

https://www.drmalpani.com/knowledge-center/promotion/why-access-to-mental-health-care-depends-on-more-than-clinical-availability

okk

Publish as is

Don't use seo optimizer

Since it is a paid article, should i publish this article as it is or use SEO optimizer

Please upload this under the Promotions folder and share the URL with me. This is a paid sponsored post/article

<media>

https://www.drmalpani.com/knowledge-center/articles/the-one-thing-your-ivf-clinic-can-actually-deliver-and-most-won-t-show-you

https://www.drmalpani.com/knowledge-center/articles/doctor-you-decide-the-three-words-that-wreck-ivf-cycles

https://www.drmalpani.com/knowledge-center/articles/why-my-ai-avatar-often-explains-ivf-better-than-i-do

https://www.drmalpani.com/knowledge-center/articles/before-you-ask-me-for-twins-go-visit-a-nicu

https://www.drmalpani.com/knowledge-center/articles/the-commonest-complication-of-ivf-is-one-nobody-calls-a-complication

Will share the update here once those are done .

AI twin bugs are already on my tasklist to fix as well as to improve a couple of things.

@analyst what are the pending tasks

Lots of glitches waiting to be fixed!

the experience of seeing you speak is very close to real but the problem is when i was speaking it only captured a couple of words and then it stopped trascribing so i was never able to speak fully or get any response

Please share an update about the fixes and the URLs

These links provided by the AI agent on the whatsapp chats are all wrong so can you please fix them as soon as possible?

I understand you're looking into ICSI: https://links.beyondchats.com/acfkr2f for low testosterone and low sperm count: https://links.beyondchats.com/acfkr2f/knowledge-center/the-infertile-man/oligospermia. ICSI can be very effective in cases of low sperm count, as we only need one healthy sperm to fertilize an egg.

There is clearly a lot of scope for improvement here

And when you tell them to email or WA the clinic, the least you can do is provide an email address and WA number they can click on

And the point is to engage the user by answering their questions and nudging them to ask the next one rather than telling them that they can email the clinic.

Where did Endometriosis come in ??

Bad answer :(

<media>

This is called the uncanny valley

<media>

<media>

Pata lag raha hai real nahi uai but you want to talk to it instead of abandoning it

Inbtween hai

Looking realistic? Ya fake?

Yaha i wanted to ask how ling have you guys in ivf treatment….

<media>

Ek baar yaha atak gaya tha then reload pe work kiya

Beech beech sei slow ho jaa raha hai randomly

<media>

https://x.com/i/status/2075807299638014212

https://x.com/i/status/2075635283211772279

Fixed costs - irrespective Variable - functin of 2 things: num of calls , total duration of calls What assumptions have we made? How do we extrapolate costs to 100 calls, 1000 calls. DOes the cost remain same? What value can we expect from the calls? How do we evaluate the calls? - did they end because all questions got answered? Did the AI make it a 2-way conversation by asking more questions, etc. or just acted like an FAQ chatbot? How could we have improved engagement, had longer conversations, What actions could we have suggested? How could we have provided more value? Eventually we need …

@analyst Analyse this group (DrMalpani webmaster)’s msgs for last 1 week. Do you have any feedback for Dr. Malpani?

@analyst Dr. Malpani needs to review Saurabh’s & Pankaj’s & Devanshu’s work in webmaster groups (DrMalpani webmaster & Internal Webmaster) and give score + feedback for this week’s performance: -Completing tasks properly: __ out of 6 -Completing tasks on time: __ out of 6 -Sharing regular updates on their own: __ out of 6 Review and share.

There are 2 charges. 1) Fixed platform fee - Heygen = $149, ElevenLabs = $22, OpenAI = $20, AWS + DB + S3 = $60 2) Variable charges - These are on a per-token basis - charged by ElevenLabs, OpenAI, HeyGen. We get a certain amount of tokens pre-included in our subscription plan. Ex: 100k credits by elevenlabs, 1000 credits by HeyGen. One this pre-included limit exceeds, each platform starts charging us on a per-credit basis. Everyone’s credit system is their own metric. We can only get a close approximation of how many tokens does 1 credit buy us. On top of this, we can only get a close appr…

I need more clarity about the AI twin costing. What does a cost of $2 for 10 min actually mean? Do the costs get incurred every time someone starts a conversation or is it a cumulative cost no matter how many conversations occur in that 10-minute window? Please provide me with a little more granularity.

And also the changes to the AI twin, for example nudging the patient to ask the next question at the end of the answer it provides to keep the conversation going

Can someone please summarize all the changes which we are going to make to the website based on Claude's suggestion?

Unfortunately, it’s limited to solving problems that I faced uptil now

It’s been a game-changer

I agree, I have automated a lot of micro-tasks and team management using analyst across various groups

https://claude.ai/share/cd807a5c-6671-4892-b9fe-cc54ec7df9b6

@102095734710467 I love this idea of using @analyst on WA. This will be great for improving communication amongst employees in a small startup because so many things start falling between the cracks

Thanks ! Let me know when the glitches are fixed.

This is already on our agenda and I am working on it already.

Also the audio conversation with an AI agent is far better than it is with us so I guess this needs to be improved

One more piece of feedback is that it picks up too much of the ambient noise as a result of which it is hard to have a conversation with the AI chatbot in public

Thanks!

Due to which users weren't able to hear the agent audio. Fixed it

It's a IOs issue

@analyst Yes

@analyst Check msgs from start of this week by Pankaj and Dr Malpani in this group. Update Tasks Tracker sheet accordingly. https://docs.google.com/spreadsheets/d/1gt7XG_9Jv5c2upZkrbMFjUn2o_fmQqTYXDqRNK2qmuM/edit?gid=683639354#gid=683639354 If the task is not yet tracked in sheet, add it in the sheet. If task is already present there, then update its status and notes. Add any links related to that task if available so we can easily check progress. Once done, share here what you did. dont die

Thanks !

Monday ko chatbot live kr denge

I had asked team to train startupmentors chatbot. But we weren’t able to crawl the website because it was not indexable properly. Aj blogs vagairah sections improve kr rhe hain.

Startup Mentors needs a chatbot too and you need to ingest a lot of the entrepreneurship and startup books from Anna's archive for the knowledge base

If you notice, the common denominator with all our efforts, whether it is education or healthcare or entrepreneurship, is to educate the end user

Model it along the lines of - everything I wish I had known when I started my startup

Excellent!

Each step will be a task or question. And each step will have additional resources - MV’s / IndianAngel’s / etc. articles, videos, etc.

@179035711746252 Please review this. I feel startupmentors should be duo-lingo style journey for young wannabe entrepreneurs

https://startupmentors.in/vision.html

Thanks so much!

https://admin-ai-twin.drmalpani.com/docs/ai-twin-costs.html AI Twin costs estimates

Yes, doing it

@267155589353519 Publish this article

Please review and implement these changes

https://claude.ai/share/46dd3fbb-183e-4706-81ab-582d5434610e

<media>

@analyst what are the pending priorities

@179035711746252 Check this AI clinic roadmap - vision doc for malpani

https://www.drmalpani.com/fertility-roadmap/roadmap.html

@analyst Check https://docs.google.com/spreadsheets/d/1gt7XG_9Jv5c2upZkrbMFjUn2o_fmQqTYXDqRNK2qmuM/edit?gid=683639354#gid=683639354 Share status of all tasks that were supposed to be done till today

Please also share the costing

Cool !

I’ll have Saurabh launch ai-twin as an embed so we can easily display it within the website.

We also need to get more traffic on the digital AI twin and the best way to do this would be to alternate it with the chatbot on the website

Excellent. And then the next step is to allow patients to access it through WA. What's the timeline for this ?

Thanks!

Using AI to move blogs into right categories.

I will also get the first version of email-analyst for patients live today.

*StartupPilot blogs section Current setup* - URL structure: /%year%/%monthnum%/%day%/%postname%/ → e.g., /2026/07/10/the-hidden-costs-of-... - ~100 published posts - 10 categories exist (building, executing, fundraising, hiring, ideas, scaling, failure, etc.) but virtually all posts are assigned to "video”. This looks like a misconfiguration - Yoast SEO is active, which is good. - No blog landing page or blog link in navigation --- The core problems: 1. Date-in-URL is the worst permalink structure for discoverability. It hurts click-through rates. It signals to both Googl…

https://www.drmalpani.com/knowledge-center/articles/nobody-throws-away-an-abandoned-baby-so-why-do-we-throw-away-embryos

https://www.drmalpani.com/knowledge-center/articles/Don-t-abandon-your-frozen-embryos

Update with AI Twin - Experimented with the new VAD filter , tuned it and integrated it into the agent for better conversations ( Optimizes and fixes issues related to agent transcriptions , agent response speed and interrupt delay with much more smoother experience ). Tested it on my end , and made it live for more edge case testing with the bot. Still actively working on the issues related to audio that might arise while talking to make the conversations as smooth as possible.

This looks like a test chat , might be done by me while testing

Yep . It is real time

No more chats with the Twin ? Is the dashboard in real time ?

<media>