Raj Agrawal

Learnings from software and technology

  • Home
  • Technology
  • Software
  • Work
  • Play

Connect

  • GitHub
  • LinkedIn
  • YouTube

Powered by Genesis

You are here: Home / Archives for Technology

Seven Ways To Predict Indian Bank Trouble From Public Data. All Seven Failed.

August 22, 2026 by Raj Agrawal Leave a Comment

Every few months a startup deck promises to predict the next bank failure from news data. I wanted to know whether that’s buildable, so one night I sat up past midnight waiting for the last of seven tests to finish. The question I’d chased all winter: can you tell which Indian bank is about to get hit, using nothing but public information? The number came in a little after one in the morning. It landed below blind guessing.

So had the other six, a couple slightly worse than random, which takes a certain talent.

The setup

Every night a monitor pulls public documents about 25 Indian banks, big and small, from every corner where bank trouble becomes public: penalties, circulars, market disclosures, court and enforcement records, cyber advisories, and news. Around 4,600 documents, each tied to its bank and linked to the original. The question: which bank gets an RBI penalty, a SEBI order, or a cyber incident in the next 14 days? Ground truth: 86 incidents (35 penalties, 16 orders, 21 cyber, plus fraud and court events). That scarcity is itself a finding.

Seven ways to ask one question

Seven ways of asking, all on the same locked split:

  • Counting: the documents, penalties, and stories trailing each bank, and whether they’re rising or falling.
  • The same counts by source.
  • The same counts by kind of event.
  • Meaning: a small open language model summarizes what the words say; simple statistics sort the summaries.
  • Word counting: which words show up, and how often.
  • Coaching: the same model, taught from a handful of worked examples.
  • Reading and judging: an AI reads the bank’s recent documents and must answer yes or no.

The pass mark isn’t fifty-fifty; for rare events it’s what a coin weighted by rarity would score, so blind guessing already banks 0.2591. My stop rule, set in advance: beat random by more than 0.05, or abandon prediction entirely. Pre-committing a kill criterion is the only defense against noise at two in the morning.

The results

MethodScoreRandom-guessing barDifference
Number-crunching the paper trail0.24330.2591−0.0158
Reading the meaning of the words0.25970.2591+0.0006
Counting the words0.25870.2591−0.0004
Coached on worked examples0.24480.2591−0.0142
An AI reading the documents0.24610.2591−0.0130

Frozen test split, 14-day head start; the bar is random guessing on the same data.

Every entry sits at or below random, bunched within two hundredths of a point. The best posted +0.0006 over random. That’s static. The best single probe managed +0.0418 (word counting, on SEBI orders), short of the 0.05 gate. The gate didn’t trip. I stopped.

Even a cheating version fails: hand a method everything said on the day the bad news broke, and it still can’t pick out the banks that got hit. A bigger model wouldn’t have helped either. Buckmann and Hill (two DeepMind researchers, arXiv 2408.03414) found simple statistics on a small model’s reading usually beat a giant model guessing from a few examples, my exact case.

Why I believe the null

Weak-signal hunts usually end with a story by slide forty. I inverted the incentives: split, yardstick, pass mark, and stop rule, all written down and locked before the final tests ran. Even my tooling refuses to publish site copy that claims prediction, a guardrail whose only job is stopping me oversell my own data. Quietly satisfying. Honest limits: 25 banks, one market, eighteen months, 86 incidents. What I can say, with receipts:

Everything a well-equipped outsider can assemble from public sources does not forecast these outcomes at a 14-day lead, and the failure is not for lack of architecture.

What I built instead

The monitor exists: a continuously updated, fully-referenced record of what’s publicly happening to each bank, free at banks.rajagrawal.com. No account, no gate. Each bank gets an attention score: how much public discussion is swirling around it right now. It ranks conversation, nothing more, and the site says so wherever the number appears. There’s a heat map, every number clicks through to its sources, and thin data gets “insufficient data” instead of a guess.

If you follow Indian banks, as an analyst, journalist, or depositor, that record is the product. If you think I’m wrong, good. The standing challenge: break the null. Find an approach, a sharper definition of trouble, or a licensed dataset that clears the bar under rules locked in advance. I want to hear about it.

Filed Under: Technology Tagged With: banking, data, india, machine-learning, rbi

Artificial Intelligence And The Art Of Killing

August 6, 2015 by Raj Agrawal Leave a Comment

Just few weeks ago, a robot killed a factory worker. The factory was that of Volkswagen, one of the worlds biggest auto manufacturers. The worker was installing the stationary robot, when it suddenly hit and crushed him. The worker died in the hospital, and now the authorities are clueless about who they should blame.

It’s said that Artificial Intelligence will be the undoing of man; that it is the hell we will create for ourselves. In this incident, the robot might have had a defect since they are not programmed to take decisions on who should live and who should die, but it serves as a perfect example for the naysayers. Everybody fears a terminator like future, where we are fighting for our very survival.

Artificial Intelligence - Asesinato de Francisco Guerrero
You don’t want a robot to be doing this. Would you? – (Analogy borrowed from ‘The Mexican Ripper’ by artist José Guadalupe Posada).

It is a rational train of thought. We create Artificial intelligence that assumes superiority over humans and assumes that we have outlived our usefulness, it launches war on us. And human kind is exterminated. I say rational not because it does not sound extremely stupid.

Elon musk thoroughly condemns Artificial intelligence. He does not cite concrete reasons for the same, but he firmly maintains the opinion. If somebody hits me by mistake while travelling on the road, I will feel a momentary surge of anger, but I will try to understand why the other person hit me. In any scenario, being the kind of person I am, I would rather not retaliate even though the other person might have done it on purpose. I want my peace of mind, and what is right and what is wrong does not concern me for such a trivial matter. Making a machine understand this is almost impossible, or at least it seems the case.
Angels and Demons; the world is made up of them. In reality it’s gray, but that’s how we generally perceive people. Darkness is evil, and the bearer of evil is the devil.

It would boil down to whether the person is right or wrong. With this in mind, one question comes to mind – how would Artificial Intelligence determine whether the person is right or wrong? One way would be to check if the person under observation has done something questionable, when compared against a list of ideal behaviours that a human being must display. Another thing to consider is the impact or his/her irregular behaviour. That is how we would think; although in our case its second nature to analyse things, so we end up not asking questions. The movies tell us that at some point in a future we may or may not be able to foresee. Artificial intelligence is going to decide that it needs to rule the world, and machines will start killing people.

Can we perhaps avoid this situation? If I make a self sustaining Artificial Intelligence machine, i’ll override some of its behaviours. First of all, there should never be a society where machines and humans co-exist as equals, simply because the purpose of us creating Artificial Intelligence is so that they can help us survive, not to establish their own civilisation. And so even if we give them the power to analyse and gather knowledge, their ultimate goal should always be that of satisfying our interests. That sounds tyrannical, but that is the only way to avoid a possible future disaster.

Paradoxes are fascinating. ‘This statement is false’ is a positive assertion which equates to true, but which contradicts the context of the statement. We can process paradoxes because we accept that two contradictory statements can exist in a similar way that contradictions exist in our daily lives. We can be good and bad at the same time, and this experience teaches us to accept conflicting views. Artificial Intelligence will merely convert statements into equations to understand them, and when it encounters something like ‘statement = false = true’, it fails. How it will react to this scenario is unpredictable.

I believe we are doing Artificial Intelligence the wrong way. Instead of getting them closer to human thinking, we should give each one of them a specialisation, and have them explore the possibilities of that specialisation. Google is already doing that with their self driven cars. I can imagine a future where a car will drive along a coast line by itself without anyone just to the enjoy the view. We can have Artificial Intelligence work on solving big world problems like pollution, poverty, crime, corruption, disease, global warming or have them working on future possibilities in Space exploration, medicine, alternate resources of fuel and food. If we try to make them exactly like human beings, we might not utilise their potential, because humans have the tendency to degrade through the way they think. Instead make them focus on one particular thing, and have them become smarter and become experts and innovators at that. This way, they might actually turn out to be a boon for society, and not the death of it as we’re predicting so conveniently now.

Filed Under: Technology Tagged With: artificial intelligence

Smartphones Have Feelings

July 4, 2015 by Raj Agrawal Leave a Comment

Decades ago, cellular phones were like obedient canines. They were faithful. They would run out of juice when you don’t feed them and would light up when they see you. When I had a simple Nokia phone, I was happy. It did not misbehave with me, and was generally predictable. Its state depended on how I treated it. I could soak it in water, and it would dry up and start working in a few hours. My uncle dropped his 3310, and it was as if something gradually grazed it; not even one bit of hurt.

A smartphone on the other hand, is like your first newborn child. It may not always be responsive and requires more juice than you can provide. But it learns like a child. These days a phone can surprise you in ways you would not imagine. If I check the score of some game played between two random teams in my phone browser, it shows me results of other games, in hope that I would appreciate it. It can automatically arrange stuff for you, and keep your memory intact. It lets you do much more than a normal phone allows you to, and has become more than just a fancy commodity you just cannot do without.

Smartphones have feelings
A smartphone that feels.

A friend of mine argues that smartness requires one to think by themselves. As such, phones being called smart is a joke. Phones only do what you tell them to. Even when they tend to do things you are not expecting, all of it has been programmed, and they are doing nothing on their own accord. Autofocus, predictive search, predictive results, behavioural understanding, and suggestions based on behavioural understanding are all part of what the phone has been designed to do. Where is the so called smartness then?

By definition, smartness is having or showing intelligence, which is the ability to acquire, understand and use knowledge. Human beings are able to process data into information, store it, and use it at a later time. It can be as simple as your wife remembering what you said 5 years ago on a boring day, at an inconsequential time, and then bringing it back in a conversation to awe and frustrate you, or it can be as complex as learning chemistry for the first time. Computers however, have to be fed data and have to be told how to use it. The only work they can do by themselves is to store the given data and retrieve it back.

Can you teach a computer how to think? This question has always boggled me. As a human being who knows that the computer is a machine, it sounds ridiculous. But as a computer programmer, it’s a world brimming with possibility. There are mainly two aspects to this, teaching a computer how to save the data in a meaningful manner based on the type of data supplied, and having it retrieve the data based on the type of query supplied. The secret ingredient in all of this is having the computer memorize this lesson of saving and retrieving data in a specific manner, and having it apply the same the next time you want the said data. So in a way, you’re incepting a behaviour in a computer’s memory. Without getting into too much detail, let’s just say that Artificial Intelligence has all of this covered. And although full fledged AI has its own separate branch, the basics of AI are used in almost all software applications these days.

The funny thing is this is exactly how we teach our children. We teach them behaviors; we just don’t have to code them in their minds. There is that extra effort involved with computers. But the bigger claim still holds; you can emulate smartness in a computer, and you can give it the ability to understand and record certain behaviours and then use them appropriately based on choices, which themselves can be programmed into its memory.

Recently, I had booked a flight from Bombay to Bangalore from a popular online booking service. They sent me an email to my gmail account. The minute I opened the mail, my phone processed it without me even having to tell it about the flight. On the day of the flight, I got an automatic reminder 3 hours prior about the pending flight. As I was reaching the airport, I got a reminder about which flight terminal the flight was leaving from, how the weather was, and if the flight was on time. In this entire process, never even once did I had to intimate anything to my phone, and it still had the courtesy of reminding me.

So suppose your context is a conversation about catching up with someone the next day at 9; if you long tap the home key, Google now will analyse the context and “smartly” ask you if you need a reminder next day at 8. Incidentally, that is also what your brain would do; register a mental reminder automatically. There are many such examples where this new feature might prove extremely handy. Technology is progressing at a rapid pace, and phones are becoming “smarter”.

Can Smartphones really have feelings? That’s akin to the ‘Android and the future of the human race’ debate. And it’s questionable to how useful will it be, if smartphones have feelings. If your smartphone could replace your friend in the future, is it a good or bad thing? In my opinion, the day man stops socialising in favour of gadgets, that will signal the end of humanity. That is one of the reasons why AI looms as a possible threat for our future. But, at the same time, I would like to end this article with the following transcript –

  • *Dial a call to Raj*
  • *Phone answers*
  • Phone – I’m sorry, but Raj is either away or busy right now. Is it okay if I give him a missed call alert and tell him to contact you later?
  • Caller – No, I need to speak with him now.
  • Phone – Oh, is it urgent?
  • Caller – Yes.
  • phone – Okay. I’ll keep ringing in order for him to notice that it might be important. At the same time i’ll message one of his colleagues and inform him to remind you.
  • Caller – Okay.
  • Phone – Is it a medical emergency?
  • Caller – No it’s something personal.
  • Phone – I’m only a phone after all. You can either leave a voice message or a text message about it. That ways he’ll see it faster.
  • Caller – No thanks. I want to speak with him personally.
  • Phone – Okay sure. Bye.

Filed Under: Technology Tagged With: intelligence

  • 1
  • 2
  • 3
  • …
  • 40
  • Next Page »