← portfolioMadam, OTP? · Project Kabootar

06 · Your Thinking Log (verbatim, in the order you worked through it)

info · Your own messages from the working sessions, quoted exactly. These are the evidence trail for how the strategy evolved: each entry shows what you asked, the constraint you set, and where you pushed back. Short one-line messages are included at the end.

1 · Setting the brief to brainstorm (10 to 15 use cases, first principles, second-order effects)

okay let's brainstorm on a few things. I have a set of insights that I've got in terms of this OTP problem. Now I have to, in a way, find a solution where voice, logistics, and payments can be used in terms of AI agents to solve these problems, but there are some constraints which I'll be giving you.

While respondents describe this opening in the language of interruption and frustration, underneath what sits is a problem of trust and identity — which is currently addressed by a six-digit code. Agents raise the stakes: trust that today passes awkwardly between humans must soon pass between humans and machines, and the OTP was never designed to carry that. The weakest solutions will simply swap the OTP for an agentic handshake and relocate the friction. The strongest will start from the emotional pain: the meeting interrupted, the mother stranded at the door and grant trust to the household as a unit, since the buyer and the receiver are so rarely the same person. Simultaneously they will need to understand and account for why platforms injected OTPs into their processes in the first place and meet their emotional needs too. Returns pickup then adds the layer that makes this opening a genuine test of the competition: one flow that must connect logistics, voice and payments to work at all.

I've written down all the insights on a web cycle and the questions that we are about to answer. Now I'm still not that convinced about how to use voice agent payments and logistics agents. What permission to give them? Which rail will need some innovation, which does not exist right now, so that this could also happen, right?

Let me also give you the judging criteria. This is where your product strategy chops are tested. You’ll have to discover a unique consumer insight around which you assemble your agent. As with most consumer products, this insight is often unexpected and surprising — sometimes it’s about asking users to hand over something they own, like their payment history, or access to their WhatsApp groups. In return, you’ll have to make the case for how your agent will fundamentally rewire their life and solve a deep, emotional problem they experience every day. Finally, you’ll have to design your agent across the three rails any agent has to be built on if it is going to work in the real world:

Every submission is read against the same five things, whichever track you enter and whether or not you write code.

now, here is what we're going to do. Let us come up with how AI agents in voice payments and logistics will solve these issues. Try to come up with 10 to 15 use casese, like different permutations, for how they are useful . They can also be separate. Let's say voice agents and payment agents and a logistics agent can only help solve the location issue, the refund issue, or the buyer-is-not-the-receiver issue, right? I will have to understand the different permutation combinations of how AI agents can be used in different ways, which can also be non-obvious. Some can be non-obvious too

but I want you to think through first principles and also think through the second-order effects, because I just don't want to give a solution and then relocate the friction somewhere else, right? Second-order effects are really important in this to understand what the consequences of this will be and which will be the most optimized way. That way, the second-order effects do not really relocate the friction but decrease it at some point and make the entire thing efficient for both people who are delivering and logistic companies, and also for buyers. It is covering a good use case. I don't want something to cover 100% of the use case, but a majority of the use case, or the Pareto principle

let's start brainstorming. Don't try to hallucinate, and don't try to add some other examples. Try to stick to the whimsical chart that I've given to you

2 · Stress-testing against real respondent quotes (#462, #326, #312)

“Coordinating drop off my online delivery. It has become an insane task to coordinate delivery for my online orders. Most orders are delivered separately and now needs an OTP. I don't live in an independent house so it needs to be delivered at the door. If I am not at home (or busy with something), my poor mother struggles with collecting otp and understanding delivery processes. Make it several times a day (because no online supermart ever has all grocery items or …”

“I had to constantly check for my Amazon delivery, since I am in a college, often the delivery guys come during lectures, so they call me for OTP or to inform that the delivery is done. I wish I had an assistant who can identify these call and coordinate the drop without disturbing me.”

“I had a return pickup fixed but the agent marked me as unavailable when I was home at all times, no call, no messages. Then I had to reply to at least 3 chatbots one asking if agent contacted, two asking if I would like to reschedule pickup and I had to reply two times to each”

Let's assume that the people who did not get the return pickup were primary because the address was not very clear on the end, and there was not a good understanding about the address, or the delivery agent was unavailable during that time. That's why he had to schedule the pickup the next day. Now he was unavailable because maybe he got delayed with other orders, etc. I'm giving you a few case problems that people have told me, quoted in their own words. The first three are the ones that they have quoted in their own words about the problems they have faced. Now, again, try to cross-check the answer that you gave above and see if they match. If the solution that you are giving will actually help people with these use cases

3 · Asking for precision: who gives the OTP, where the agent lives, what the framework is

Can you be more precise in terms of this OTP thing? Who is giving the OTP? Where will the agent work? Will there be an agent on the receiver's phone? Does that mean that all the people should get the agent, install the agent, or that the agent will be on the logistics people's phone or the delivery guy's phone? How will that work?

Can you be more precise about the steps about the agent? Try to answer all of these permutations that you gave in the clusters, right in a particular framework, and follow this framework when it comes to:

I want to understand all of this in this format for all the clusters that you gave earlier, in terms of how the OTP will work and how the agent will work.

4 · The shoe-return quote (#85)

“I ordered a pair of shoes online, but the size was wrong. I had to find the order, check the return deadline, keep the original box and packaging, choose a pickup slot, and make sure someone would be home. The pickup did not happen on the first day, so I had to open the app again and reschedule it. Each step was small, but together it felt like too much work for one return.”

What about this problem? How can you solve this problem using voice and using logistic agents?

5 · Fraud in a low-trust society

But we are also forgetting that in India, a lot of fraud problems will happen because, again, all these fraud problems are there. What if the buyer then just takes the box in this case and just goes away with it? In reality, there are a lot of fraud problems that can happen because of this. How do we solve for that using agents also? India is a low trust society. We have to take that into consideration.

6 · Track: Product Strategy. Mind map and founder-level understanding of the rails

build a mindmap/flow chart for the all the use cases for both power/ core and casual users and lets not just say that the existing technology doens't have any gofence at the moment but lets undestand on how to then find a solution for this. what test can we do or mvp (theoritical and design mocks up)can we build using their existing api information that we know.

i want to also get a founder level understanding of the technology rails : voice, logistics and payment agents in this context and get a foundational understanding of this so that i can think of an innovative solution by myself from grounds up

7 · System design thinking: shadow inference layer, barge-in, LLM, sentiment, Gnani languages

This means anyone building on top of a logistics API has to build what I'd call a shadow inference layer — using sparse, delayed checkpoints to statistically estimate things the API doesn't tell you directly. ETA prediction from historical route-and-pincode data is a classic example: the API never tells you "arriving in 8 minutes," but if you've logged enough past deliveries on that route, you can infer it. This is actually a legitimate, defensible moat to build, because it requires accumulated proprietary data, not just an API integration — worth thinking about as your own innovation surfac

think of yourself as a system designer and a developer. what could be the system design of building this (a mvp) which can be feasible and low effort.

lives in the dialog logic and failure handling — what happens when the person interrupts mid-sentence (barge-in), what happens on silence, how quickly the system decides "this isn't working, fall back to a human or a text message." Most voice-agent startups that fail, fail on this orchestration layer, not on the underlying speech models. If you want to invent something original here, look at interruption handling and graceful degradation, not at "better transcription."

for our use case will we use an external llm? and how can we do the sentiment analysis. what posisble logic can we use? for that and also try to understand how good gnani is for multi languages?

also use second order thinking to then further undestand the consequences of these product decisions

8 · Autonomy, time saved, trust and safety, phased roadmap

  1. Define the Autonomy and Trigger
  1. Prove Value via "Time Saved" and Efficiency
  1. Outline the Trust and Safety Architecture
  1. Propose a Phased Risk-Mitigation Roadmap

can we also deep dive into these questions and further use second order effects to find a better solution

9 · Why the rider call and always-on GPS cannot be the trigger

The rider's own call as the arrival trigger the problem with that is that usually these delivery people don't even call, and sometimes they fake the call. They call, but they're not at the location, and they just tell us that they're not able to deliver. That is the problem, right? Sometimes these people are lazy, or they don't want to do it, and they just don't call. We cannot take that as a proxy that they are at the location. They might lie about it. For the physical scan, the idea was to be near the scanner, and the location should be on so that it knows that it's near the house, so that there are fewer chances of fraud happening. But we can also eliminate the geofence thing while scanning it. The physical scan and the doorstep would necessarily not need GPS, but GPS would definitely be needed for the ETA because we want to calculate the live ETA of the rider to the desired person's house. Like how you have in Swiggy and Zomato, you have a system where you can see the person's location in real time. We don't want to do that in real time, but we want to have some time stuff about it, some idea.

10 · Fraud apart from geofence; one QR across many orders

Apart from the geo fence thing, what other things, or in any other way, do you think we can solve for scams in this way, or for frauds? The QR thing: the problem with the QR thing is that I'm not able to figure it out. If there are, let's say, 10 orders, multiple orders, how do you then make a single QR that could be approved for all the orders? Let's say it is from Delhivery, or if you get 3 parcels from Delhivery and 4 parcels from Shadowbox, but they are coming individually, maybe in a span of 3-Can we think of some other way or some other solution apart from geofence? 4 days, let's say 2 - 2-4 or , something like that. Are you then solving for the QR thing as a design thing? Do you make a single QR? How do you send it? At the same time, make sure that the fraud happens where the delivery guy is taking a photo of the QR and not really scanning it, etc. can we think of some other solution apart from geo fence which is also techinically feasible solution to add and can be integerated with the exisiting capabilities of the delivery and gannin

11 · Universal rotating QR (Cult.fit example)

just tell me how can we build a universal qr that lets say will get updated in every 15-20 mins and will stay with just the reciever so that he or she can use it for differnt orders from the same delivery aggregator without having the fear or fraud. cult for eg has a qr sytem which is unique for every user but unique for every gym. they are solving it

12 · QR and voice for leave-at-door, 10 to 15 orders, and pickups with nobody home

Listen I want you to think about how we will use QR and voice agents for two use cases: 1. How will we use QR if, let's say, the guy is telling the delivery guy, "Just keep my order near the door"? Let's say he lives in an apartment or something. No one is in the house, he lives alone, and he wants the delivery guy to keep the order near the door, maybe on the shoe rack or something. Here how will we solve for QR and voice agents? 2. Let us say there are 15 orders or 10 orders. For these 10 orders how will QR be solved for? Also let's say there is pickup. For pickup assume that, again, the guy kept the pickup box outside his house. Now how will QR solve for that? In pickups you have to give an OTP. How will we solve for that? Assume no one is in the house, no one is assigned, and there's only one person who's living alone. How will we solve for that?

13 · Operator questions and outreach format

  1. is the delivery route optimized for the delivery agent?

  2. verify if for COD payment otp is needed or not?

  3. how is NDR logged in general? is there any place to give reasons by the delivery agent for NDR?

  4. why isn't some sort of tracking and estimated ETA feature has not been rolled out yet? inside logistics aggregator companies

Based on this, can you come up with a LinkedIn post and also a post that I have to post on a community channel? Try to follow this format, where you have a hook, then you have a context, and then you have these questions that we are asking. I have given a format below. Just refer to this format

Hook: Start with "I know that you might probably be busy but I just wanted your advice" or "I just wanted to ask you something" and then ask the question. Ask only one question, then give context:

CONTEXT: I'm actually working for this competition under the KEN, where we are learning on how to build AI agents for delihvery. We just wanted to get context about what has already been worked on so we don't repeat the same mistakes and we get more of an idea about how the company operates internally

Try to come up with one format that I could use to cold text people on LinkedIn, but make sure this only follows 200 characters. Try to ask the second question because it's a low-effort question, and then after that, I'll see how to ask the questions.

Come up with the LinkedIn post so that we can have outbound people who can then contact us, and also the community post that I'm posting. Please make sure the format is followed.

14 · Short messages (verbatim)

"but do u think the delivery ride will always keep his gps on? does he usually does that?"

"i dont want the user to do anything. he should not do any work. such feature don't work where the user needs to do any sort of work"

"lets think about this new hypotheis adding an ip address could be a crazy addition to the existing qr"

Respondent quotes used as evidence (verbatim from the problem statement)

"Coordinating drop off my online delivery. It has become an insane task to coordinate delivery for my online orders. Most orders are delivered separately and now needs an OTP. I don't live in an independent house so it needs to be delivered at the door. If I am not at home (or busy with something), my poor mother struggles with collecting otp and understanding delivery processes. Make it several times a day (because no online supermart ever has all grocery items or …" — Respondent #462, 25–34, Bengaluru

"I had to constantly check for my Amazon delivery, since I am in a college, often the delivery guys come during lectures, so they call me for OTP or to inform that the delivery is done. I wish I had an assistant who can identify these call and coordinate the drop without disturbing me." — Respondent #326, 25–34, Mumbai

"I had a return pickup fixed but the agent marked me as unavailable when I was home at all times, no call, no messages. Then I had to reply to at least 3 chatbots one asking if agent contacted, two asking if I would like to reschedule pickup and I had to reply two times to each" — Respondent #312, 18–24, Indore

"Sharing my location with delivery partners, UC staff and the laptop repair guy despite giving it on the app." — Respondent #399, 25–34, Delhi

"I ordered a pair of shoes online, but the size was wrong. I had to find the order, check the return deadline, keep the original box and packaging, choose a pickup slot, and make sure someone would be home. The pickup did not happen on the first day, so I had to open the app again and reschedule it. Each step was small, but together it felt like too much work for one return." — Respondent #85, 18–24, Bhopal

"i have ordered some 8-9 items from amazon by my friend's account and i was calling him regularly 4-5 days to ask him which parcel will come today which was quite hectic as he was calling me sometimes that parcel has arrived but i was somewhere out and wasn't available" — Respondent #151, 18–24, Hyderabad

Back to 00 README Index