The real opportunity of AI is to make women’s healthcare proactive instead of reactive. AI can help interpret symptoms, identify risks early, personalize guidance, and support clinicians in delivering care at a scale that traditional systems simply cannot. Akanksha Vyas, CTO & Co-founder, Pinky Promise shares exclusively with Rajneesh De, Group Editor, APAC Media & CXO Media that from a technical perspective what excites her most is the opportunity to build clinical AI systems that work for women.
Pinky Promise is an AI-enabled, chat-first clinic. What differentiates your platform from generic health chatbots, especially in handling complex and nuanced women’s health cases?
The most important thing to understand about how Pinky Promise works is that a doctor is at the centre of every consultation, and the AI is built to augment the capacity of doctors rather than to replace them. Generic health chatbots typically provide informational guidance or symptom suggestions. They surface information and suggest next steps, but do not carry clinical responsibility for any of it.
What we have built is fundamentally different in its design. Our AI handles the parts of a consultation that consume enormous clinical time without requiring clinical judgment, gathering symptoms in a structured way, organising the patient’s history, drafting a differential diagnosis and a prescription for the doctor to review. The doctor then examines all of that, makes their own clinical call, and every single piece of medical information the patient receives always comes from the Doctor.
The other distinction that matters is that we did not just build on top of a general-purpose medical LLM. We recognised very early that women’s health cannot be treated as a subset of general medicine. The conditions, the stigma, the way symptoms present, the way patients describe them, all of this is specific enough to require models built and trained specifically for this domain.
We have built custom AI models trained on over 10,000 sources including 250+ clinical protocols, peer-reviewed literature, and contributions from senior gynaecologists, as well as our own data that we anonymize. We also use LLMs as a language layer to personalise the experience of each patient, but the clinical intelligence underneath is ours, built with purpose.
Pinky Promise serves women across every district in India, spanning geographies, languages, and health literacy levels. How do you design for that range without compromising care quality?
There are two problems folded into this question and they require a different kind of thinking. The first is linguistic and the second is clinical. On the linguistic side, the reality of how Indian users communicate digitally is that they do not neatly separate languages. Someone typing about a health concern on her phone in Jaipur or Patna is likely typing in a mix of Hindi and English, using Roman script for both, switching mid-sentence without thinking about it. Our platform is built to handle that as a given rather than an edge case. We currently support Hinglish and English and we are quickly expanding from there to support other languages.
Pinky Promise deals with sexual and reproductive healthcare conditions, which are ripe with stigma and have different euphemisms for many important terms that differ from district to district. We have pressure tested our medical questions with over 10,000 women across India to make sure the language and wording we use is simple and clear across the board.
The clinical quality question is where the architecture and design matters most. We have designed the UI of the system to make it very easy and intuitive for a user to describe their symptoms and conditions. Moreover, as the AI assists the Doctor in asking the patient questions about their symptoms and is trained on the latest scientific evidence and protocols, it ensures that no question is missed. The doctor always receives a consistent, well-organised and complete clinical summary.
A patient with low health literacy and a patient who has already researched her condition extensively are both presented to the doctor through the same structured framework. That means the quality of the clinical picture the doctor works from does not depend on how articulate or medically literate the user was. We have also built a RAG-based copilot that learns the individual communication style of each doctor and of each patient on the platform. The responses retain a personal, contextual quality and communicate with the patient in the language and style she is comfortable with.
With a large volume of consultations and a high rate of cases resolved on-platform, what does the data reveal about women’s health-seeking behaviour that traditional care settings often miss?
The finding that has shaped almost every product decision we have made is the honesty gap. We discovered it early, before we wrote a single line of code. Divya was running experiments where she mimicked a chatbot on Facebook and WhatsApp groups. She began asking the same medical questions in text and then following up by phone to ask them again.
Women were consistently more forthcoming on chat rather than on the call. They disclosed things they had not mentioned in the verbal conversation, symptoms they considered embarrassing, medications they had taken, concerns they had been carrying privately for a long time. That gap is not a minor one. It is the difference between a doctor having the full clinical picture and making the right call, and working from partial information and potentially missing something important. Patients also tend to trust the diagnoses more, because they have been honest.
We hear from our doctors often that the kind of cases they see on Pinky Promise are very different from cases they see in the clinics. Traditional Gynaecology is typically focused around pregnancy. But the reality is that women face sexual and reproductive issues, often chronic in nature, from puberty to menopause.
Period irregularity, vaginal discharge, contraception, what to do when a cycle does not arrive when expected, these are all standard questions that often have clear answers. The fact that women have been carrying them privately rather than asking a doctor is a function of how care has historically been delivered, and our data makes that visible in a way that is difficult to ignore.
What the platform data also reveals is the scale of first-contact users. Roughly 53% of the women who come to us have never consulted a gynaecologist before, and these are adult women, many managing conditions like PCOS or recurrent infections over years without any professional input. Our three-month retention rate of 97.4% also tells us something significant, which is that once a woman experiences care that feels genuinely safe, she comes back.
As Co-founder and CTO, you are building AI systems for a highly sensitive space. What were the toughest technical decisions you had to make while designing the platform’s core architecture?
The toughest architectural decision was deciding where AI should stop. One of our earliest architectural decisions was separating language intelligence from clinical reasoning. Large language models are incredibly effective at understanding messy, multilingual patient conversations, and communicating with empathy in the language of the patient. They also have incredible reasoning, but they make mistakes. Healthcare demands consistency, protocol adherence, and accountability. We designed Pinky Promise to bring together the best of both. The AI assists the Doctor with each of the steps, but final clinical decisions are always clinician-led.
Another difficult architectural decision was designing for escalation, not automation. In healthcare, a system should not optimize only for speed, it must know when to pause, ask more, or escalate to urgent care. From the beginning we have built pathways for risk flags, ambiguity, severe symptoms, and cases requiring urgent offline care and escalate them immediately.
Finally, a core decision we have made from day one is architecting for privacy. We store all patient data in-house and no patient data ever leaves our servers. Any data used in model training is fully anonymised with personally identifiable and medical information stripped out before it is touched. Patient records are encrypted at rest using AES-256 and in transit using TLS 1.2 protocols, and only the consulting doctor has access to a patient’s file. In a space where the reason women come to us is precisely because they need to feel safe, there is no version of the product that works without it.
As a technology leader in healthtech, how do you ensure that AI augments clinical decision-making rather than replacing critical human judgment?
The way the system is designed, the AI augments and enhances the capacity of the Doctors, but never replaces them. The AI helps with everything that is structured and tedious for the doctor – gathering symptoms, organising the patient’s history, running through differential diagnoses, preparing a draft prescription, so that by the time the doctor reviews the case she is not spending her limited time on intake and administration. She is spending it on the clinical thinking that requires her training and judgment.
A good gynaecologist sitting in a clinic can see 30 to 40 patients a day. With our model, she can care for 10x of this number without any reduction in the quality of her clinical engagement, because the cognitive load on the repetitive parts of each consultation has been absorbed by the system.
What we have been very deliberate about is where the hard boundary sits. Not a single line of medical information reaches the patient without a doctor having reviewed and approved it. This is not a policy statement we enforce through process. It is a hard architectural constraint in how the platform is built. The AI can draft, organise, and suggest. It cannot prescribe or speak to the patient directly. That boundary exists because medical accuracy is of paramount importance. Trust in a medical product is extraordinarily difficult to build and very easy to lose, and we are not willing to trade long-term trust for short-term efficiency.
You come from medtech, fintech, and a 3D-printing health startup before Pinky Promise. How has that cross-domain background shaped how you approach building AI systems for women’s health?
Coming into women’s healthcare from multiple industries gave me one big advantage: I wasn’t constrained by how healthcare has traditionally been delivered. Pinky Promise is my second venture as a founder and my fourth time as a founding CTO of a startup, each experience has added something specific to how I think about building.
At Fited, my medical 3D printing company, we built software to mass-customize and build 3D-print Scoliosis braces. We built and tested the product with hospitals in the Netherlands and Turkey, and what I really learned from this experience was that no matter how cool or cutting edge the technology is, what matters at the end is if the patient’s symptoms improve or not.
From work in Fintech and legal tech at Idfy and OpenLaw, I carried over a rigour around trust infrastructure. Financial products and health products share a fundamental design challenge, which is that users are giving you something they cannot afford to have mishandled. Building for security or privacy is paramount. The moment the system loses their trust, it is extremely hard to rebuild. That shaped how seriously we approach every architectural decision that touches user data or clinical safety.
I joined Pinky Promise after trying out a very early version of the platform myself. It was a condition I had been self-medicating for fifteen years and received a correct diagnosis in ten minutes. That is when I understood that Pinky Promise is truly different from all other digital health platforms I have encountered, and I needed to be a part of building this.
What is your long-term vision for AI-led women’s healthcare, and what role do you see yourself playing in shaping that ecosystem?
When I was growing up, my mother rarely went to a doctor and I am often the same way. We tend to be very reactive about our own healthcare, and only prioritize it when the condition becomes urgent. Today we have the opportunity to change this for the next generation. The future I imagine is one where a woman’s first interaction with healthcare is as simple as opening a chat and being understood instantly, in her own language, without judgment, from anywhere, even at 2am. We need to make healthcare as easy as getting groceries or a loan is today.
However, the real opportunity is to make women’s healthcare proactive instead of reactive. AI can help interpret symptoms, identify risks early, personalize guidance, and support clinicians in delivering care at a scale that traditional systems simply cannot. The opportunity in AI-led care is to move upstream, to personalise health guidance based on an individual’s longitudinal profile, predict risks before they become serious, and support ongoing management of chronic concerns in a way that does not require a woman to keep returning to a clinic every time she needs input.
We have already started building in that direction with our 24×7 chat-first approach, long-term digital care programmes, and predictive health AI frameworks. We have already seen very strong early outcomes, including women conceiving after years of managing undiagnosed endometriosis, early detection of cervical cancer, etc.
From a technical perspective what excites me most is the opportunity to build clinical AI systems that work for women. There is a huge gap in published women’s health data. In fact, women were not allowed in clinical trials till as late as 1993. Most of the clinical AI tools available today were trained on data that is predominantly male, and the consequences of that bias show up in how women are diagnosed and treated across virtually every medical domain.
FemTech platforms are in a position to correct for this, but only if they approach data as a clinical asset built with care and specificity rather than volume alone. Beyond serving the patient, we want to build tools for providers and for the wider healthcare ecosystem, so that women’s health finally has the data infrastructure and clinical intelligence it has always deserved.
