# Karthick Aravindan > I build software and AI products, explore financial systems, study how businesses work, and document what I learn along the way. Site statement: Building with intelligence. Thinking in systems. This is a personal notebook. Published notes have a date and a body. Drafts are omitted from this file, the sitemap, search, and the public writing index. Finance notes are a personal viewpoint. They are not investment advice, not a recommendation to buy or sell anything, and not a report of results. ## Start here - [Home](https://karthickaravindan.com/): Building with intelligence. Thinking in systems. - [About](https://karthickaravindan.com/about): Background in software engineering, and the current lines of work. - [Work](https://karthickaravindan.com/work): Selected products and trading systems. - [Writing](https://karthickaravindan.com/writing): Published essays on AI, finance, and building. - [Principles](https://karthickaravindan.com/principles): Principles being tested, with the evidence kept separate from the claim. - [Timeline](https://karthickaravindan.com/timeline): Public project record. - [Books](https://karthickaravindan.com/books): Books read, with a personal note on each. - [FAQ](https://karthickaravindan.com/faq): Short answers about the person, the notebook, and how to get in touch. - [Search](https://karthickaravindan.com/search): Search published essays, books, principles, projects, influences, and tools. - [Contact](https://karthickaravindan.com/contact): Contact form and a public 30-minute call link. ## Projects - Focus20 Labs (Launched): Building practical AI solutions, agents, and workflow automation for real business problems. https://focus20labs.com/ - Restro20 (Launched): A restaurant POS for everyday operational workflows. https://restro20.com/ - GrabSaaS (Launched): A public site for finding SaaS products, offers, and tools aimed at founders and small teams. https://grabsaas.com/ - Own Research (Launched): Analytics and simulation tools. https://ownresearch.in/ - Finance and Trading Systems (In Progress): Building the trading systems and algorithmic market research for Treasure Capital. https://treasurecapital.in/ ## Published notes - [Why I Prefer Hedged Option-Selling Strategies Over Unhedged Directional Bets](https://karthickaravindan.com/writing/why-i-prefer-hedged-option-selling): Why I want the loss of a position understood before I enter, and why option selling, option buying, and intraday trading are not labels that settle the question. Published 2026-10-26. - [Why I Build So Many Things](https://karthickaravindan.com/writing/why-i-build-so-many-things): On the tension between many experiments and the few products that show real demand. No invented outcomes. Published 2026-10-26. - [Systems Over Motivation](https://karthickaravindan.com/writing/systems-over-motivation): On Atomic Habits and how I try to apply it to learning, building, and review. The book's ideas and my experience are kept apart. Published 2026-10-26. - [Building to Learn](https://karthickaravindan.com/writing/building-to-learn): On prototypes, the Lean Startup, and the limit of what I can know before something runs. Published 2026-10-26. - [First Principles in Product Development](https://karthickaravindan.com/writing/first-principles-in-product-development): On decomposing a problem, questioning the usual solution, and reading Zero to One without turning it into a guarantee. Published 2026-10-26. - [What Trading Taught Me About Systems and Risk](https://karthickaravindan.com/writing/what-trading-taught-me-about-systems): On probability, position structure, backtests, and why I will not judge a strategy by a short winning streak. No private rules. Published 2026-10-26. - [Building AI products beyond the demo](https://karthickaravindan.com/writing/building-ai-products-beyond-the-demo): On the distance between a demo that looks finished and a product that holds up in use. Published 2026-10-26. - [Why distribution is harder than building](https://karthickaravindan.com/writing/why-distribution-is-harder-than-building): On finding the people a product is for, after the building has started. Published 2026-10-26. - [What backtesting can and cannot tell you](https://karthickaravindan.com/writing/what-backtesting-can-and-cannot-tell-you): On the limits of historical tests when designing a systematic trading process. Published 2026-10-26. - [Lessons from building products for real businesses](https://karthickaravindan.com/writing/lessons-from-building-products-for-real-businesses): On what changes when a product has to survive an actual business workflow. Published 2026-10-26. - [Thinking about risk as an engineer](https://karthickaravindan.com/writing/thinking-about-risk-as-an-engineer): On treating risk as part of the system design, in products and in markets. Published 2026-10-26. ## Machine-readable - [Sitemap](https://karthickaravindan.com/sitemap.xml) - [RSS](https://karthickaravindan.com/feed.xml) - [Full text for models](https://karthickaravindan.com/llms-full.txt) # Published note text ## Why I Prefer Hedged Option-Selling Strategies Over Unhedged Directional Bets URL: https://karthickaravindan.com/writing/why-i-prefer-hedged-option-selling Published: 2026-10-26 Why I want the loss of a position understood before I enter, and why option selling, option buying, and intraday trading are not labels that settle the question. This is a working preference, not a result. I would rather hold a hedged option-selling structure than an unhedged directional bet, because I want the loss described before I enter. I am writing that preference down as a hypothesis about how I want to take risk. It is open to evidence. It is not a finding that this family of trades wins, and it is not a suggestion that anyone else should trade it. A view about direction and a position are different things. The view is a story about where price might go. The position is the payoff I actually hold if that story is early, late, or wrong. Two people can share the story and still lose in different ways. I care about the position. ## Understanding the loss before I enter Before I enter, I want a plain account of the loss. What happens if price rises, if it falls, and if it goes nowhere? How large is the loss if I am slightly wrong, and if I am badly wrong? Can I get out, and what kind of market do I need in order to do that? How many of these positions can sit together before one bad day is no longer acceptable? A plan to pay attention leaves that account unwritten. So does a limit I have not specified. I use the same standard I use when I build a system. I understand it when I can say how it fails, how large the failure can be, and what I do then. That is a personal standard. Meeting it does not, by itself, produce a higher return. ## Directional speculation and a structured strategy Directional speculation, as I am using the words, is a position whose payoff depends on the market moving a particular way, with the loss left open or left to an exit I hope to make in time. A purchased call that needs a fast rise, an uncovered short put that needs price to stay above a strike, and an unhedged futures position are different instruments. What they share is that the story about direction is doing work the structure has not done. A structured, risk-managed strategy names more than the story. It names the conditions for entry, the legs, the size, the hedge, the exit, and the loss I am willing to realize. It can still lean in a direction. A short put spread prefers the underlying to remain above a region of prices. Structure is a translation of a view into a payoff I can describe when the view fails. It is not a claim that I have no view, and it is not a claim that the view is correct. ## Why hedging and defined risk matter to me A hedge, in the sense I want, is a payoff that improves in the scenario where the primary position is losing. A second trade that does nothing in that scenario is another risk. The structure I keep coming back to is a vertical credit spread. I sell one option and buy another in the same expiration, further out of the money, on the side where the short option can be hurt. For a short call, that is a higher-strike call. For a short put, that is a lower-strike put. In the textbook case, where both legs are still on, the loss levels off near the difference between the strikes minus the net credit per unit, multiplied by the contract size. Fees sit on top of that. The long option costs money. That cost is the price of the cap. I count it inside the strategy. I use the word "defined" narrowly. That textbook figure holds when the short option has not been assigned into a different exposure, and when I can still get out of both legs. On contracts that allow early exercise, the short option can be assigned before expiration. I can be left long or short the underlying while the other option is still open. Until that is closed, it is a different position. A thin market can also make the exit worse than the diagram. The contract decides which of these apply: the multiplier, how it settles, and whether early exercise exists. I am describing the general mechanics. I am not restating one exchange's rulebook, and I am not publishing the contracts I trade. An uncovered short option has no such cap. Smaller size reduces the damage and leaves the shape of the payoff unchanged. Size and a hedge do different jobs. I want the size chosen before the position is on, and I want the hedge to be a payoff, not a mood. The essay page carries a sketch with invented numbers, so three shapes are easy to see. A long option, whose loss levels off around the premium. An uncovered short call, whose loss does not level off as price rises. A defined-risk spread, whose loss flattens. The numbers are made up. The sketch is not a backtest, not an account, and not a trade. ## Time decay, implied volatility, timing, liquidity, and cost ### Theta Theta is the change in an option's price as time passes, with the other pricing inputs held still. The contract's theta is usually negative. Extrinsic value tends to fall as expiration approaches, if nothing else changes. The buyer of the option feels that decay as a headwind. The seller, who is short the same contract, has the opposite exposure, so the passage of time can help the short position, again if nothing else changes. When people say a short option "has positive theta," they mean the position. They do not mean that the contract's price rises merely because a day went by. Theta is not a coupon. It changes with the time remaining, with how close the strike is to the underlying, and with volatility. In the usual pricing models, the decay of an at-the-money option often becomes a larger share of the premium that remains as expiration nears. That is also when a short option's sensitivity to a price move can become sharp. The decay and the risk arrive together. I do not treat theta as evidence that selling is a favorable thing to do. ### Implied volatility Implied volatility is the volatility number that, inside a pricing model, reproduces the option's market price. It is a way of quoting that price. It is not a forecast I get to treat as a fact about the future. When implied volatility is higher, the option is more expensive, all else equal. The seller collects a larger credit. The buyer pays more for the same contractual right. After the trade, the comparisons that matter are the movement that was priced against the movement that occurs, and the implied volatility I traded against implied volatility later. A short option can lose money because implied volatility rises, even when the underlying has barely moved. A long option can lose money because implied volatility falls, even when the direction was right. I am not publishing a rule for when volatility is rich or cheap. The pricing relationship is standard. A trading rule about it would stay a hypothesis until it was tested on data that had not been used to invent the rule. ### Entry timing The same strikes are a different trade if I enter them before a known event, into a wide quote, or after implied volatility has already moved. Timing is part of the position. Conviction about direction is a weak substitute for it. I am not writing an entry rule in this note. ### Liquidity A diagram has a line. A market has a bid and an ask, and sometimes neither is useful. The distance between them is a cost, whatever I believe about direction. Volume and open interest are signs that other people have traded the contract. They are not a promise that a usable quote will be there on the day I most want one. A hedge I cannot close or adjust at a tolerable price is thinner than the chart that persuaded me to put it on. ### Transaction costs Commissions, exchange fees, and the spread I have to cross belong in the expected result. A spread has at least two legs, so I pay to enter and leave both of them. An adjustment adds further trades. A path computed from mid prices, with no slippage, describes a market I do not get to trade. Slippage is the difference between the price I assumed and the price I receive. It tends to be worse when I am in a hurry. I am not putting a number on it here. A description of a strategy that leaves these costs out is incomplete. ## Selling options is not automatically safer or more profitable Selling an option collects a premium in exchange for an obligation. The premium is the price of that obligation. It is not a yield. Collecting it is not evidence that the trade has a positive expected value. A short option can expire worthless often enough that the win rate looks comfortable. A high win rate can still sit next to a negative expected value when the losses are large. That sentence is arithmetic. It is not a statistic from my trading, and it is not a claim that short options, in general, win often. Short options are also short gamma. As the underlying moves against the position, the exposure tends to grow, so the loss can speed up rather than stay in a straight line. Implied volatility can rise. Price can gap through a strike. Positions I thought were separate can be hit by the same move. A margin requirement can force a reduction at a poor price. What a broker will actually do depends on the account agreement and the contract. The distinction I rely on is simpler. A margin rule is not a contractual cap on the loss. I do not believe that selling options, as a category, raises the chance of profit, improves the expected return, or makes a strategy safer in every market. An uncovered short option can be more dangerous than a purchased option, because the seller's loss can be far larger than the credit. The version I am willing to study is conditional. The loss has to be capped or otherwise bounded. The size has to make that bounded loss acceptable. The credit, the cost of the hedge, and the cost of trading all have to be inside the sum. Those are conditions I would require. They are not a property of the word "selling." ## Uncovered short calls An uncovered short call obligates the seller to sell the underlying at the strike if assigned. Where the contract is cash-settled, the seller pays the cash equivalent of that obligation instead. Either way, there is no theoretical upper bound on the underlying's price. As that price rises, the cost of meeting the obligation can rise without a ceiling. The premium offsets only a finite amount of it. This is the contractual shape of the position. It is not a forecast about any market. A stop order does not install that ceiling. Price can gap through the stop. A margin requirement can demand more capital or force an exit, and the exit can happen at a price the diagram never marked. I do not want a structure whose worst case I cannot state. ## Uncovered short puts An uncovered short put obligates the seller to buy the underlying at the strike if assigned. A cash-settled contract makes the seller pay the cash equivalent. The shape is different from the short call. Where the underlying cannot trade below zero, the loss has a theoretical bound: if the price goes to zero, the loss approaches the strike value of the contract, minus the credit received, before fees. That bound can still be a severe loss beside the credit, on a high-priced underlying or a large position. Zero is not a universal floor. Some commodity futures have settled below zero, so the instrument decides how far the tail goes. I am not going to call a short put safe because a stock is familiar, or because the loss has a number attached to it. A large bounded loss is still a loss. Cash set aside to secure the put is the cash that buys the underlying if assignment comes. It names whose money is at stake. It does not remove the loss. Where the contract allows early assignment, the purchase can be forced before expiration. I would size that loss and accept it on purpose. Receiving the credit is not the same act as accepting it. ## Buying options has a real place A purchased option is a right. If I keep the position as an option — I hold it, I sell it to close, or I let it expire — the most I can lose on that option is generally the premium I paid, excluding commissions and other fees. The premium can go to zero. That is a total loss of what I paid, and a series of those losses can become expensive. The loss on the option itself still has a boundary. Exercise is a separate decision. Exercising a call means paying the strike and taking on the underlying, which can then fall. Exercising a put disposes of the underlying, and it can leave a short position in the underlying, which has its own risk. The premium limit belongs to the option. It does not travel with the new position. Buying options is a reasonable category of risk when the terms are accepted with open eyes. The long option is what makes a credit spread's loss finite. A long option can also express a view when I want a known maximum loss, and when I accept what that purchase requires: the underlying has to move far enough, and soon enough, to overcome the premium, the time decay, and any decline in implied volatility. The habit I would question is buying short-dated options because each individual loss feels small. Small and repeated is still a process. It needs the same kind of evidence I would demand of anything I sell. ## Intraday is a time horizon Intraday means the position is opened and closed inside the same trading day. That is a holding period. It does not say whether the position is long or short, whether the loss is capped, or whether the idea has an edge. Risk controls and evidence matter more to me than the label. An intraday credit spread and a credit spread held for several weeks can both be sized, costed, and tested. An intraday uncovered short option is not a milder trade because it will be closed before the session ends. If that option is also close to expiration, its sensitivity to price can be extreme, and one session can contain a large share of the loss. A short holding period also multiplies the number of times I cross the spread, and the number of chances I have to put on a trade without a fresh reason. A historical test of an intraday idea that ignores slippage, fees, and the gap between a stored price and a live quote is not yet evidence. A short horizon has an ordinary failure mode: the session keeps offering another entry. Leverage can turn a small error of judgment into a large error in the account. I am not claiming that every intraday method fails. I am claiming that the clock does not supply the strategy. I have read about short-term methods, including How to Make Money in Intraday Trading, by Ashwani Gujral and Rachana A. Vaidya. The book's claims stay with the book. I am not taking a rule from it here, and I have not recorded which printing I read. ## Tests I would want before trusting this A backtest answers a narrow question. It says how a stated rule would have behaved under the assumptions that were coded: the fills, the costs, the period, the universe, and the hours included. Change those assumptions and the result can change. I want them written down with the result. Out-of-sample validation is the next requirement. A rule that is only shown on the history used to design it has been compared with the memory of the design. I want a period, a market, or a walk-forward split that was not used to choose the parameters. Slippage belongs inside the test, with commissions and the bid-ask spread. Stress is a separate pass: gaps, a jump in volatility, a stretch where positions I treated as independent lose together, a session in which the hedge is hard to trade, and a change in margin. A smooth path through ordinary weeks does not answer those cases. Position size is part of the strategy. A defined loss on one spread can still be too large once it is multiplied by the number of contracts. I am not publishing a sizing formula. The standard I want is plain. If the loss I call defined would impair the account, then the definition lived on the diagram and the position was something larger. I have not published results for any of this. Own Research is where I work on analytics and simulation tools. [Treasure Capital](https://treasurecapital.in/) is where I am building trading systems and algorithmic market research. Those sentences describe the work. They are not a return, a win rate, or a set of rules. ## What would make me reconsider I would revise this preference if the evidence changed, or if the market conditions that give the preference its shape were absent. A louder argument would leave it where it is. The evidence I mean is a clearly specified alternative. It might be option buying, an unhedged directional position, an intraday method, or another structure. I would want results that hold up out of sample, after realistic costs and slippage, across more than one regime, together with a stress test that shows the bad path. I would want the comparison to include how large the losses get, alongside how often the trades finish ahead. I would also revise the preference if hedged structures of the kind I prefer failed those same tests. A view that cannot lose an argument with the data is a view I should not keep. Market conditions could pressure it as well. If implied volatility stayed cheap relative to the movement that followed, the credit would be poor pay for being short gamma, and selling would be a weak place to start. A long run of gaps and one-way moves can make even a defined-risk short-premium structure a poor way to take risk, because the path of the losses still matters after each loss has a cap. If the options I can actually trade became too wide, or too hard to leave under stress, the hedge would exist on the diagram and not in the market. A change in exercise, settlement, or margin practice could also pull the live loss away from the textbook loss. I would reconsider, too, if the hedge consumed so much of the premium that a simpler position with its own hard boundary had the clearer expectancy after costs. A long option is one such position. A small position in the underlying, with a loss that is pre-committed and can actually be executed, is another. A hedge is worth what it covers, at the price it charges. ## Where this leaves me I am more interested in the structure of a position than in predicting every move. Hedged option selling is the structure I currently prefer, because I want the failure described before I am in it. That preference is a hypothesis. It is not an edge I am reporting, not a description of an account, and not a ranking in which selling options always finishes ahead. Buying options has a place in that hypothesis, including as the hedge that puts a boundary on a short option. An uncovered short call can lose without a theoretical ceiling. An uncovered short put can lose much more than the credit, up to a bound that can still be severe. Intraday trading stays available as a horizon when a tested process supports the trade. The name of the horizon does not do that work. P. R. Sundar is one of the people whose public options commentary I follow. The public work treats a position as something that is constructed and then adjusted, which is the question I care about. I want to understand the approach and then test the assumptions myself. I am not attributing a setup, a rule, or a quotation to him. Following the work does not mean I share every view he holds, and it does not mean he has any connection to me. The Complete Guide to Option Selling, by James Cordier and Michael Gross, is part of the reading behind the subject. I am not borrowing the book's claims about results, and I have not recorded which edition I read. ## Why I Build So Many Things URL: https://karthickaravindan.com/writing/why-i-build-so-many-things Published: 2026-10-26 On the tension between many experiments and the few products that show real demand. No invented outcomes. I build often. The public record is a restaurant POS, a site for finding SaaS products, analytics and simulation tools, an AI practice, research that is still in progress, and a set of ideas that never became products. I do not experience that list as a single company. I experience it as a way of finding out what I think, by making the thing. Curiosity is a good reason to start. It is a bad reason to keep funding every start with the same attention. The Lean Startup gave me a language for that: an idea is a hypothesis, and the build is not the test. The test is what someone does when the thing exists. I have not published the numbers that would let a reader score these projects. I am not going to invent customers, revenue, or a winner. The tension I actually feel is between range and concentration. Range teaches me. I learn how a restaurant workflow differs from a research tool, and how an AI workflow differs from both. Concentration is what I owe to anything that starts to show demand. Demand, for me, has to be behaviour I can point at. Interest from friends, or my own excitement, does not count as that evidence. Distribution is the part I underestimate while I am building. Making the software can feel like the whole problem because it fills the day. Finding the people it is for, and learning whether they stay, is a second project. Several of the things I have shipped are public. Being public is not the same as being chosen. Some entries in the timeline are still ideas, or still unfinished. ReEngage20 is in progress. FixedSLM is a concept, not a service. I would rather leave them in that state than promote them into successes. Stopping, or leaving an idea as an idea, is part of the same principle as shipping. The point of many experiments is to earn the right to focus, not to keep a museum of half-used domains. What I want from the next year of building is a shorter list. Not because range was a mistake, but because a notebook full of products is not a substitute for one product whose use I can describe without bluffing. ## Systems Over Motivation URL: https://karthickaravindan.com/writing/systems-over-motivation Published: 2026-10-26 On Atomic Habits and how I try to apply it to learning, building, and review. The book's ideas and my experience are kept apart. Atomic Habits, by James Clear, argues that small repeated behaviours, and the environment around them, matter more than waiting to feel motivated. That is the book's idea. What follows is how I am trying to use it. I have not logged a habit streak, and I am not claiming a personal transformation. Motivation is real and unreliable. I notice it most when I am starting something new. It is a poor operating plan for the fifth week, when the work is ordinary. A system is the ordinary version: a time I already protect, a next action small enough to start, and a note I write when I stop so the next session is not a search. In technical learning, the system I trust is a loop, not a syllabus I feel guilty about. Read a small piece. Make it run. Write one sentence about what broke. The sentence is the part I skip when I am only consuming. Without it, I confuse familiarity with skill. In product work, the system is a review, not a burst of features. What did a person try to do, and where did the software get in the way? If I cannot answer from observation, I am adding scope to soothe myself. James Clear's framing would call that a habit of motion. I would call it staying busy. They are close, and neither one is an improvement. I do not think a one-percent gain compounds into a guaranteed outcome. Costs, a bad base process, and simple forgetting can wipe the gain out. The idea I want to keep is narrower. If a change is not inside a cycle I repeat, it is not a system. It is an event. The book is not my diary. Where my experience is thin, I would rather say so than invent a morning routine to match the chapter. ## Building to Learn URL: https://karthickaravindan.com/writing/building-to-learn Published: 2026-10-26 On prototypes, the Lean Startup, and the limit of what I can know before something runs. I trust a small working thing more than a long explanation of the thing I have not built. The explanation can be tidy and still omit the constraint that shows up in the first afternoon of implementation. Building is how I meet that constraint. Eric Ries describes a loop of building, measuring, and learning, with the product idea held as a hypothesis. I use that as a discipline, not as a ceremony. The useful version is short. What do I believe will happen? What did I put in front of someone, or in front of a dataset? What happened that I did not predict? A prototype is allowed to be incomplete. It is not allowed to be incapable of being wrong. If every result can be narrated as progress, I did not run an experiment. I ran a demonstration. This is also how I want to treat trading research. A simulator or a backtest is a build. It teaches me what my assumptions were. It does not, by existing, prove that a strategy makes money. Own Research is my public place for analytics and simulation tools. I am not publishing a result from it in this note. Focus20 Labs is where I build practical AI systems for real workflows. The same rule applies. A workflow that only works in a demo has not been learned yet. It has been performed. The limit is cost. I should not build a large system to answer a question a conversation or a spreadsheet could answer. Building to learn is a choice about the cheapest contact with reality, not a reason to write more code. ## First Principles in Product Development URL: https://karthickaravindan.com/writing/first-principles-in-product-development Published: 2026-10-26 On decomposing a problem, questioning the usual solution, and reading Zero to One without turning it into a guarantee. First principles, as I mean it here, is a way of slowing down at the start. What is the job? What is actually constrained — time, money, regulation, physics, a person's attention? Which parts of the usual product are there because they solve that job, and which are there because the last product in the category had them? I associate the public version of this habit with engineers who talk about reasoning from constraints, and with product work that refuses features which do not change the experience. Elon Musk and Steve Jobs are two of the people I study for that. I am not borrowing a quotation, and I am not saying their companies are a template for mine. Zero to One, by Peter Thiel with Blake Masters, pushes a related question: are you making something new, or competing as a variation? I think the question is worth sitting with. I do not think the book is a machine that produces a successful company. A clever distinction can still be aimed at a problem nobody has. In practice I try to write three sentences before I design a screen. Who is stuck? What do they do today? What would have to be true for a new tool to be worth the switch? Restro20 started from everyday restaurant operations, not from a desire to have a POS in a portfolio. Focus20 Labs starts from a business workflow, not from a model I want to demo. Those are intentions. They are not case studies with outcomes attached. The failure mode I watch for is using "first principles" as a performance. If the decomposition does not change a decision, it was a preface. The point is the constraint I am willing to design around, and the assumption I am willing to drop. ## What Trading Taught Me About Systems and Risk URL: https://karthickaravindan.com/writing/what-trading-taught-me-about-systems Published: 2026-10-26 On probability, position structure, backtests, and why I will not judge a strategy by a short winning streak. No private rules. Trading is the clearest place I have found where a story about being right can outrun the system that is supposed to hold the risk. A few winning trades feel like understanding. They are often a sample. The system question is what happens across the trades I do not get to choose in hindsight, including the ones that gap. I care about the position more than the prediction. Probability is not a mood. A high chance of a small gain can sit next to a real chance of a loss I cannot repair. Position size is part of the strategy. So is the decision about whether the loss is defined by the structure or only by my intention to exit in time. Backtesting is useful and easy to flatter myself with. It tells me how a rule would have behaved under the assumptions I coded: fills, costs, the period I chose, the universe I chose. Change those and the story changes. A test that never leaves the data I used to design it is not confirmation. Out-of-sample checks and ugly scenarios are the minimum I want before I take a result seriously. I have not published those results. Own Research is where I work on analytics and simulation tools. [Treasure Capital](https://treasurecapital.in/) is where I am building the trading systems and the algorithmic research. I am not describing the rules, the sizing, or a return. Public notes can talk about the questions. They should not leak a private playbook. P. R. Sundar's public work is one of the places I watch how options positions are discussed. I then want my own test. Agreement is not the goal. A reason I can inspect is the goal. What trading keeps teaching me, outside markets, is that a system is defined by its bad day. Software, a business, a research loop: if I only describe it when it works, I do not understand it yet. ## Building AI products beyond the demo URL: https://karthickaravindan.com/writing/building-ai-products-beyond-the-demo Published: 2026-10-26 On the distance between a demo that looks finished and a product that holds up in use. A demo can look finished. Someone asks a question, the model answers, and the room treats that as the product. I have stopped trusting that moment. The interesting gap is between an answer that looks right once and a system that still works on a Tuesday, inside someone else's workflow, when the input is messy and nobody is presenting it. Focus20 Labs is where I work on practical AI: agents, workflow automation, and tools aimed at real business problems. The public site is the place to see the practice. I am not publishing a case study here, and I am not claiming a customer outcome. The lesson I can state is smaller. A demo proves that a path exists. It does not prove that the path survives contact with the work. What usually breaks is not the sentence the model produced. It is the surroundings. Who is allowed to act on the answer. What happens when the source is missing. How a person corrects a wrong step without starting over. Whether the system leaves a record. Those are product questions. They do not show up in a highlight reel. I would rather ship a narrow workflow that a person can rely on than a broad demonstration that only works while I am in the room. The demo is a start. The product is what remains after the demo is over. ## Why distribution is harder than building URL: https://karthickaravindan.com/writing/why-distribution-is-harder-than-building Published: 2026-10-26 On finding the people a product is for, after the building has started. Building fills the day, so it is easy to mistake it for the whole problem. Distribution is the other problem: finding the people a product is for, and learning whether they stay. I underestimate it while the software is still taking shape. I have shipped public products. Restro20 is a restaurant POS. GrabSaaS is a site for finding SaaS products, offers, and tools. Own Research is analytics and simulation tools. Focus20 Labs is the AI practice. Being on the internet is not the same as being chosen. I have not published conversion numbers or revenue, and I am not going to invent them to make this note sound finished. The hard part is that the build gives constant feedback and distribution often gives silence. A feature can be completed. A person either changes how they work or they do not. Praise is not that change. A visit is not that change. I still think the build matters. You cannot distribute a thing that does not exist. I no longer think the build answers the question of who it is for. That question has to be asked on purpose, and the answer has to be behaviour, not my own excitement. ## What backtesting can and cannot tell you URL: https://karthickaravindan.com/writing/what-backtesting-can-and-cannot-tell-you Published: 2026-10-26 On the limits of historical tests when designing a systematic trading process. A backtest tells me how a rule would have behaved under the assumptions I coded. Fills, costs, the period, the universe, the hours I decided to include. Change any of those and the story can change. That is useful. It is a way to see my own assumptions. It is not a promise about the next year. What a historical test cannot tell me is whether I will follow the rule when it hurts, whether the market I tested is the market I will trade, or whether a result I tuned until it looked kind will survive data I did not use to design it. A test that never leaves the sample I learned on is not confirmation. It is a memory of the design process. I want the ugly cases in the test: gaps, costs, a stretch where the idea does nothing, a stretch where losses cluster. I also want a check on data I did not use to invent the rule. I have not published those results. Own Research is where I work on analytics and simulation tools. [Treasure Capital](https://treasurecapital.in/) is where I am building the trading systems and the algorithmic research. Neither sentence is a performance record. A backtest can reject a bad idea. It cannot, by itself, crown a good one. ## Lessons from building products for real businesses URL: https://karthickaravindan.com/writing/lessons-from-building-products-for-real-businesses Published: 2026-10-26 On what changes when a product has to survive an actual business workflow. A product for a real business has to survive a workflow that already exists. People have a way of taking an order, closing a day, finding a tool, or getting an answer. The software is a guest in that sequence. If it only makes sense in a diagram, it will be worked around. Restro20 started from restaurant operations, not from a wish to own a POS. GrabSaaS is a public site aimed at founders and small teams looking for software. Focus20 Labs is aimed at business workflows, not at a model I want to demonstrate. Those are the intentions. I have not published a study of how those products changed anyone's day, so I will not write one here. What changes, compared with a personal project, is the cost of a small miss. A confusing step is not a quirk. It is time taken from the work the business already had to do. The lesson I trust is to name the job first: who is stuck, what they do today, and what would have to be true for a new tool to be worth the switch. I am still learning that. Shipping is not the lesson. The lesson is whether the workflow is simpler after the software is in it. ## Thinking about risk as an engineer URL: https://karthickaravindan.com/writing/thinking-about-risk-as-an-engineer Published: 2026-10-26 On treating risk as part of the system design, in products and in markets. I came to markets with an engineer's habit. A system is not understood when I can describe the happy path. It is understood when I can say how it fails, how large the failure can be, and what I do then. That applies to software and to a position. In a product, risk is a bad input, a missing permission, a step that cannot be undone. In a trade, risk is the payoff when the view is wrong, late, or right in a way that does not pay. Win rate is a weak summary in both places. A process can succeed often and still be badly designed if the failure is unbounded or unexamined. I want the loss defined before I am impressed by the gain. A limit written down is different from a plan to pay attention. A backtest is a question about assumptions, not a trophy. I am building trading systems and algorithmic research for [Treasure Capital](https://treasurecapital.in/), and I use Own Research for analytics and simulation. I am not publishing the rules or a result. The same sentence covers both kinds of work. If I only describe the system when it works, I do not understand it yet.