With love, Rajat_
Engineer, Creative, Tinkerer
My name is Rajat, and my fascination with building started when I was a kid and my brother showed me how to make simple games on our computer. I remember thinking "that’s, cool!" That moment of watching code become play opened a door I've been walking through ever since, from games to systems that genuinely help people.
Today, I build AI products that do real work. At SRSWTI, I designed a production multi-agent framework that cut manual workloads by 55% for pilot clients in insurance and logistics. I've architected systems processing 200,000+ records daily, served AI orchestration APIs with 99.9% uptime, and built the kind of infrastructure that quietly runs in the background while people get their jobs done faster. The metrics look good on paper, but what actually matters to me is harder to measure: that moment when someone uses what I've built and feels like somebody actually gave a damn about their experience.
I believe what we make stands testament to who we are. In a world obsessed with what's easily counted—speed, cost, scale—I stay focused on what isn't: delight, care, humanity. I'm not interested in breaking things for the sake of innovation; I'm interested in building something so much better that the old way naturally becomes obsolete.
Outside of work, you'll find me on long walks, in the gym, or lost in music and photography. Books and architecture train my eye; conversations with interesting people shape my heart.
mémoire
mémoire enables any LLM to remember conversations, learn from interactions, and maintain context across sessions.
- • Creations stand testament to who we are, our values and preoccupations.
- • Design should genuinely move humanity forward, not just meet price points.
- • Goal: enable and inspire people as builders in service of humanity.
- • Innovation is not about breaking things deliberately, only break as consequence of creating something better.
- • Progress requires conviction, vision, and resolve; it's not inevitable.
- • "Sincerely elevate the species"—millions engage with every detail, even cable packaging.
- • Users should feel "somebody gave a damn about me" when using your product.
- • Creating products is expressing gratitude to the species.
- • Reframe quality vs speed: ask "how can we work efficiently to create breathtaking quality?"
- • Words used to frame problems shape how we think.
- • Unintended consequences are inevitable, hope they're pleasant, but responsibility is mandatory.
- • Today's rate of change is too fast for society to create proper frameworks, unlike the Industrial Revolution.
We habituate to everything except the things we love. Those stay forever new.
Habit is a survival feature: once something is familiar and safe, your nervous system stops devoting full attention to it.
Love is the exception because it's not just pleasure, it's ongoing relevance.
When you really love something or someone, your brain tags it as: "this matters to my future. Keep watching."
That's evolution's hack to keep you paying attention to your kids, your partner, your work—whatever actually matters. Everything else fades into the background.
That's why people who truly love their craft become unstoppable. The work stays forever novel and interesting to them, while everyone else burns out.
- • Simplicity is not minimalism, it's expressing essence, purpose, and role in life.
- • Removing clutter without intention creates desiccated, soulless products.
- • Joy and humor are essential, not trivial; they've been missing in tech.
- • Your mental state while designing embodies the final work—anxiety creates anxious products.
- • The measurable trap: teams default to numbers (cost, speed, weight) because it's inclusive.
- • Much of what designers contribute cannot be measured numerically.
- • Delightful, joyful products are equally important and get used more.
- • Utility and aesthetics are not opposed—if something doesn't work, it's ugly.
- • When choosing between options, the more humane one is always correct.
- • Users sense both care and carelessness—most companies patronize consumers.
- • The inside matters: finish the back of drawers even if unseen, evolution is measured by what we do when no one watches.
- • Every business has an obligation to care about design, not just consumer electronics.
LA Shelter Platform
A platform to connect LA residents affected by fires with temporary housing options.
Currently
- • AI Software Engineer at SRSWTI building production multi-agent frameworks
- • Building a personal OS on the side
- • Graduated with Masters from University of Texas
- • Running a moodboard Instagram page
Previously
- • Shipped an AI healthcare insurance app for Fidelity Investments
- • Launched a platform for listing available shelters and resources to help those displaced by the 2025 LA fires
- • Software Engineer at WeafTech, worked on data platform, distributed systems and machine learning projects with some very incredible engineers
- • Interned at TUV Nord and was part of digital transformation initiative for their industrial inspection division
- • Studied Computer Science at Birla Institute of Technology and Science
- • Launched a bite-sized news app at 18
- • Get good at something you actually love, because only love beats habituation.
- • Innovation serves progress, not disruption for its own sake.
- • Don't let old people screw you with student debt and fake prestige.
- • Exercise everyday.
- • Taste is a muscle.
- • Sweat the details.
- • A lot of things matter a little, a few things matter a lot.
- • Don't repeat talking points — you will live a boring, unfree life.
- • Face your fears — the worst outcome is death and that's coming anyway.
- • Embarrassment is the dumbest fear — literally nobody cares.
- • If motivations remain constant, you'll find ways to maintain care and control.
- • Work is important, but if you have nothing you love outside work, your brain becomes unbalanced and brittle.
- • Great outfits should be repeated.
- • The cabinet maker principle: finish what no one sees—superficiality creates nagging emptiness.
- • All work should express love and gratitude to the species.
- • Act with urgency.
- • Do what you say you will.
- • Show up on time.
There is a beautiful, largely unexplored space opening up right now: the transformation of human knowledge, sensors, and actuators from a "Human-First" orientation to an "LLM-First" one.
We are currently stuck in a primitive loop of "PDF-to-Text." We feed models raw streams of tokens and ask them to predict the next word top-to-bottom. Frankly, that’s lame. It’s the computational equivalent of passive reading.
Total LLMification
I’ve become obsessed with a better way: Total LLMification. Imagine a physics textbook. For a human, it’s a static PDF. But for an LLM, that same book should be a dynamic, functional environment. This isn't just OCR; it's a non-trivial, human-in-the-loop transformation that turns "reading" into "doing."
The Blueprint for LLM-Legible Knowledge
To truly upgrade an LLM's capability, we need to stop treating data as flat text and start treating it as structured training signals. Here is the topology for the perfect textbook transformation:
-
The Exposition (Context Layer)
All narrative text is extracted into clean Markdown. We preserve the semantic structure—tables, lists, bolding, and italics—and render all math in LaTeX. Figures are extracted not just as pixels, but as distinct image entities tied to their context.
-
Worked Problems (SFT Signal)
Every "Example Problem" in the book is converted into a Supervised Fine-Tuning (SFT) example. We parse references to previous figures or tables so the model learns how to retrieve context, not just hallucinate answers.
-
Practice Problems (RL Environments)
This is where it gets interesting. We don't just feed the model the practice problems; we extract them into environment examples for Reinforcement Learning (RL).
- The problem becomes the state.
- The answer key becomes the reward function.
- Additional context is structured as an "Answer Key" for a potential LLM Judge to score the reasoning.
-
The Infinity Generator (Synthetic
Expansion)
This is the killer app. For every specific problem type, we don't just teach the model the answer; we build a Python-based generator that creates infinite variations of that problem.
Take a simple question: "What is the angle between the hour and minute hands at 9:00 a.m.?"
Instead of learning that one fact, we generalize the logic into code. We can now generate a dataset of millions of unique time-angle problems:
- 11:07 a.m. $\rightarrow$ 68°
- 4:14 a.m. $\rightarrow$ 43°
- 8:37 p.m. $\rightarrow$ 36°
The LLM as a Student
By indexing this data into a RAG database or serving it via MCP (Model Context Protocol) servers, we fundamentally change the learning process.
Just as a human student takes a high school physics course—reading the theory, watching the teacher solve examples, and then grinding through practice problems—an LLM can now "take" the course in the exact same way.
Flashrpc
rpc over websockets made easy, robust, and production ready. a fast and durable bidirectional json rpc channel over websockets.
- • Small teams who trust, love, and care about each other are fundamental.
- • Caring means actually listening, most ideas die because people just want to speak.
- • Opinions aren't ideas, greatest fear is missing amazing ideas from quiet people.
- • Make things for each other daily: creates vulnerability, gratitude, and shifts focus to others.
What if I told you we've been looking at information all wrong? We are living inside a symphony of data, yet we've become fixated on the individual notes, completely missing the music itself. The question now is: what are we going to do about it? Can you actually see the wave functions without becoming part of the wave?
The Velocity of Knowing
"Boy logic" tells us that a wave function is just a property of a particle—something used to judge momentum, time, position, or spin. But information acts like a particle, too. It doesn't just exist; it carries context, nuance, and unexplored connections to other explanations.
Think about velocity. In physics, it's speed with direction. In information theory, there is a "velocity of knowing." The information regarding how propulsion works carries a different velocity for a rocket scientist than it does for someone working at Joe & The Juice. Both might inherit the same raw data, but the speed and direction of their understanding are radically different.
The Failure of Brute Force
Walk into any hospital, research lab, or engineering floor, and you will see brilliant minds connecting invisible dots. They aren't doing this through rigid taxonomies or explicit links. A doctor doesn't need a database query to see a pattern in patient history; they do it through a natural weaving of understanding.
Yet, here we are in 2025, still trying to brute-force these connections. We have trapped ourselves in a binary of bad options: either we throw millions of dollars at domain experts who cannot scale, or we throw billions at Large Language Models that miss the elegant simplicity of human intuition. We've built a world where true understanding comes with either a prohibitive price tag or a computational overhead that would make Moore's Law blush.
A Natural Topology
But what if there was another way? What if, instead of forcing connections through artificial means, we let information find its own topology? Think of it like water finding its natural path, or proteins folding into their perfect form.
We need to embrace:
- The dance between explicit and implicit connections.
- The natural hierarchy of understanding.
- The deterministic patterns that emerge when we stop forcing and start observing.
It sounds impossible, but graph theory and hierarchical determinism can reveal what has been hiding in plain sight.
The Math Behind the Magic
At its core, this approach recognizes three fundamental layers of information topology. We define the strength of the connection between any two nodes ($n$) at a layer using a weighted relationship:
$$\omega_{ij}^{\prime}=\alpha S_{ij}+\beta E_{ij}+\gamma K_{ij}$$
Where:
- $S_{ij}$ represents semantic similarity.
- $E_{ij}$ captures the entity overlap ratio.
- $\kappa_{ij}$ measures key phrase resonance.
- $\alpha$ and $\beta$ are layer-specific weights that adapt to the complexity of the content.
The beauty emerges in how these layers interact. A node's effective state isn't static; it's a summation of its neighborhood and base state:
$$\psi_{n}=\overline{\sum_{i\in k}}\sum_{j\in N_{i}\langle n\rangle}\omega_{n,j}^{i}\phi_{j}$$
Consider a medical case study mentioning "patient response to treatment A." In our topology, this single node naturally belongs to the treatment methodology cluster (semantic), the patient outcomes group (entity relationship), and multiple temporal sequences (contextual) all at once.
To measure this, we look at the Information Radius (IRad), a metric that helps us understand how information naturally clusters and flows without us forcing artificial boundaries.
From Static to Spectral
When we move beyond static connections, things get interesting. Because we can't standardize the "information velocity" for everyone (the engineer vs. the staff worker), we have to personalize the communities.
Borrowing from concepts I learned regarding spectral filtering and spectral transformers (nod to E. Hazan and Ana Choromanska), we can map this flow:
- Input Sequence: Start with data points $xt$ (words, sensor readings, control inputs).
- Impulse Response: Compute how the system reacts over time.
- Hankel Matrix: Use outer products of the impulse response to form a Hankel Matrix ($Z$), capturing the sequence dynamics.
- Eigen Decomposition: Extract spectral filters from $Z$ to reveal the hidden structure.
