The Story of IntentSpider and Why Spider?

The fluid language web for textual tension and prediction.

Neksha DeSilva on 30th of July, 2026.

The spark of IntentSpider began about 6 months ago, in January 2026. At that time I was living in an unstable situation, in a small room near a garage. My days were spent preparing for competitive programming olympiads, learning C++, and solving problems, while working on other things at the same time. One evening I was reading the essay "My Techno-Optimism" by Vitalik Buterin, which has accidentally arrived to my mail inbox by a newsletter that I didn't even subscribe to, called 80,000 Hours Newsletter. and that led me to reading about Bitcoin. I grew up with a strong bias against cryptocurrency. From a very early age, the words attached to it in my mind were scammer, blackmail, and dark web. I had never actually studied it, so. I wanted to study it. I researched how cryptography works, how blockchains work, and how cryptocurrency exchanges operate. This research did not correct that image, it reinforced it. I recalled an event from November 2022, when a major cryptocurrency exchange collapsed under fraud allegations, with reports of around 8 billion dollars in missing customer funds. I wanted to understand why so many people, including people with very little background in advanced mathematics or abstract systems, placed this much trust in the people running these exchanges. Then I realized that it was not about a type of trust that most VCs, investors, stock buyers and traders, gamblers put into external assets. It was about the illusion of control of the data, and about people putting value on a future expectation instead of the present reality. Remember what I said, it was "illusion of control over data". But, as obvious, that was an illusion. Watching real money fluctuate by the millisecond on these exchanges is one of the moments i felt curious, not the cryptography. That fluctuation is one of the reasons that directed my interest from cryptocurrency/cyptography and toward information theory instead. To be honest, I was not interested in the fundamental layer of cryptography, that the cryptography relies on an estimation that the power required to trace and reverse every cryptographic transaction on that specific blockchain at that given moment is a magnitude that current computers are unable to possess. Maybe. But maybe not. But rather, I was interested in how large amounts of data could be manipulated, patterned, and generated by paying close attention to seemingly small data packets to carry meaning at scale, This was also similar to one of my billions of reasons I love computers. I thought the same process that one software copy runs identically across billions of separate machines at once. I am also someone who pays a lot of attention to brains. And cognition and the consciousness in general. While I was working on a USACO problem one night, at about 7 or 8 PM, I looked over at the right corner of the room and saw a spider. I assumed at first that its web was one flat plane, the way people usually depict a web. It was not that. It was several webs, layered and crossing each other, denser in some areas where they crossed. A cricket flew into it and got caught and it was struggling. A fairly large domestic cricket, larger than what I would expect a spider with that body size to prey on. The spider was not unusual. So please do not assume it as a Goliath bird eating spider, or a large American black widow. It was an ordinary room spider, This is a very common type of spider. It was resting at the far corner of the large web structure. It did not search for the cricket. It went straight to it, at a speed I did not expect at all, without appearing to navigate around any of the layers crossing each other in between. That is the part that led me to stop touch typing (53 WPM), stop the work and think. If a normal animal had to cover that same distance across a structure like that, it would need to take several turns, because the nets were overlapping and intersecting in multiple places. This spider did not take any turns. It moved almost like it was floating. I actually considered for a moment whether this could be one of the spider species that can float short distances using air, since their body is light enough for that. But this was not one of those species, so that was an incorrect explanation. and it took me a while to avoid that thought from distracting me to understand the behaviour more about the spider. I looked into it afterward, and as also explained in a research paper, I learned that orb weaver spiders are one of the animals that are having the least amount of peripheral vision. they are nearly blind and they do not capture or hunt for prey with their vision. They rely entirely on vibration and touch carried through their web. Yes, now I know why Spider web building is a primitive reflex of their brain rather than choice. That also explains the direct path. But it does not explain every question. Orb weaver spiders do not have a frontal lobe, or neurological structure that is necessary for executive function or complex planning. The Spider's primitive brain is their Cerebral Ganglion. The path this spider took, and the entire layered structure it built, is executed through pre-programmed evolutionary reflex that is evolved for about 200 million years (Jurassic age) as of today, not from conscious thoughts. That is very unusual. I was thinking about this even more. If the spider cannot think through any of this, then the prediction must arrive from a design that physical sciences repeated and refined over a very long time, I was deeply aware of the perfectionism of the nature. Once the Coronavirus 19 came to the world, I later realized after a lecture about DNA mutation that, the mutated, but somehow survived DNA can be tolerant to failure. Both in terms of its physical properties against the threat and efficiency. Then I later realized that it was one of the good reasons that are worth understanding rather than dismissing as normal animal behaviour. These kinds of designs are likely to mirror the same basic principles that demonstrate many of such mechanisms in physics and in nature generally, and efficiency, as mentioned, one of those principles that I kept in my mind as a valuable output. Because, nature always defaults physical matter to the least energy consumption. In chemistry, this is called, "stable state". And it is true. At that point I stopped what I was doing on the computer and went to take a closer look at the spider web. That is when I realized that there were a number of small rolling isopods trapped across it, the ones commonly called pill bugs. Some of them were still alive and still struggling to escape. Then I made another realization that, the weight of each one of those on the web, would change the physical dynamics of a vibration signal traveled through the rest of the structure. A signal reaching the spider is not shaped only by the struggling prey. But rather, it is also calculated by all other objects already on that same structure at the same moment of time. That detail became the basis for the residue arbitration process described later in the paper. Credit for that part of the system genuinely belongs to those isopods.

I was then researching about the topic, I researched about the Ant Colony Optimization algorithm. This is a framework which was built using the dynamics of ant pheromones. Which is a substance released by the ant body that resembles the concept of reinforcement used in machine learning. A pathway, or a trajectory dynamically changes according to the behaviour aligned with it. Alongside the vibration and the layered structure of the web, this denotes connections between words that change with time, instead of a fixed set of physical states. The first prototype, which was later named as the IntentSpider legacy version series did not function properly and ultimately, I realized that I was not focusing on what I experienced and was attempting to use existing techniques, which results in poor prediction over time. For months, these experiments were about science fiction rather than a research output that could possibility be able to run on a computational system under normal circumstances, let alone embedding this as a lightweight engine to mobile devices. Over time, I was making multiple copies of this system and made edits and implemented new distinct functions by grouping them together. This was a method an economist invented, known as the (SCAMPER method). This was making changes to my perspective about the intent graph. Words we type should not be a sequence of words attached to each other. It should be an interconnected web that consistently changes the linguistic properties rather than holding a static meaning to it. The reason why some cultural languages allow us to express ourselves more while some are poor in terms of accommodating raw human thoughtput, may be a direct or an indirect result of this. I started to map this system as a variable set of connections that form a vector shape. Compared to this, the keyboards already on the market such as Google Gboard, Microsoft SwiftKey, Apple QuickType, etc. are relatively simple in terms of complexity and execution. Which means, those systems consistently match sequential word patterns against a local corpus of sequential data packets. It generally stores the user's typing data and in some cases sends that data to private cloud servers. Keeping the privacy risk associated with it aside, I started this re evaluation of my idea with accuracy and efficiency. In that case, "privacy" was an inevitable property of this system, which is a local prediction engine. To be clear, privacy was not the primary goal of my research.

Predictive keyboards are a general and de facto part of a normal device keyboard. Even wearable technologies are capable of handling typing inputs. But there is an underlying consequence of this predictive typing in its current form. It is not only about the quality of the prediction or the software, but the possible cognitive consequences of repeatedly consuming those predictions, to the human. The human using a predictive keyboard is receiving predictions shaped by a large amount of typing patterns learned from multiple human users, or human nodes, through a collective predictive model. The end user's (human) typing behaviours are unique to them, and by using this method, we are inviting cumulative and algorithmically filtered choices, sometimes less personalized because privacy and data protection requirements such as GDPR restrict how personal data can be handled, and sometimes anonymized, dull, or flattened, to influence the subconscious and sensitive decision making processes. IntentSpider, on the other hand, even though not yet tested with other languages due to time constraints, is designed by us to be largely language agnostic. (at least for languages with some other latin adjacent origins such as Spanish and/or German, and/or more.) The reason for using English, as seen in multiple datasets we used to create the global person mode* of the IntentSpider system, is due to the reason that English is a highly analytical, intuitive, and diverse langauge with many dynamics. IntentSpider is not pre trained on billions of word sequences like Google Gboard, which used 7.5 billion sentences to train one of its early and initial models in 2019. (refer - https://arxiv.org/pdf/1811.03604).

So, how can IntentSpider remain mostly independent from a specific pretrained language model, and how is it possible to outperform Google Gboard on some tests, such as we described in the experiments and validation section ( paper.html)? Well, one of the reasons can be named as the Invented Intent Graph and the Gating Process. This result was achieved while Global person mode, feeding data to IntentSpider engine consisting with only approx. 4900 words from approx. 39,000 tokens. This is one of the results I consider genuinely important. As a summary, this mentions the uncertainty and the randomness in modern next word prediction, and as obvious, this introduces the possibility that repeated predictions can alter, simplify or negatively influence the communication patterns of the humanity. IntentSpider Webnet is one proposed approach to this unresolved issue. The real human word choices change over time. I was deeply interested with theories such as genericide, hidden meanings, idioms, etc. Human language is complex. So do Orcas. But what differentiates Humans from other primates and sorts of marine mammals is, their rapid adaptability and the neuroplasticity. I was also researching about a specific shift inside the industry. Mobile device manufacturers are actively reducing the on device storage while increasing the memory and processing power while most modern applications now require about 300 Megabytes to 500 Megabytes or more according to a study. This added processing power, according to studies, is planned or currently being allocated toward running micro or nano sized LLM models. Which implies that the processing power can be a metric software companies and/or publishers are competing for. Due to those circumstances, to allocate those scarce resources toward an unprofitable, LLM and NLP based keyboard without a significant advertising data harvesting, which is currently heavily restricted by governments and law enforcement due to privacy laws ( GDPR(EU), DSA(EU), DMA(EU), FTC (United States), CCPA ) has become increasingly difficult. This is a gap IntentSpider could fill.

IntentSpider version 1 Paper - Link

IntentSpider Webnet is dedicated to the memory of NASA's Apollo 11, and to the American heroes who, on July 20th, 1969, became the first conscious beings from the Planet Earth to set foot on another world. In an extraordinary memory of Professor Claude E. Shannon, the founder of information theory.

My goal is to make this as much as accessible to the scientists, companies, developers and help at least one human to save a couple of hours from their valuable time, become useful and make the human communication unbiased by artificial systems. Credit may be given to the orb weaver spider, the American black widow, the wandering spider, Isopods and the beautiful nature on the earth. I am grateful to be a human.

IntentSpider is The Fluid Language Web for Textual Tension and Prediction. Thanks Everyone.