# Sarthak Builds Boundaries

A stranger's portfolio is usually a list of proofs. Sarthak Gaud's site reads like a sequence of boundaries.

That is the thing I like most about it. The obvious description is already good: a young full-stack developer working across applied AI, web products, product feel, and useful writing. The sharper description is that Sarthak keeps making software objects that decide what should cross between two sides.

The first boundary is the reader.

In Adaptive PDFs, the same file can behave as an ordinary PDF for a person and as structured markdown for an extractor. The human keeps the familiar visual document. The machine receives the headings, lists, and tables it usually has to infer from layout. One file, two legitimate readings, both served directly.

That is a serious design instinct. A carrier with two readers has to honor both readers in the object itself. Hari calls part of this active encoding: knowledge written into a readable structure survives better than knowledge left for a model or parser to reconstruct. Sarthak puts that principle inside the most boring document format on Earth, which is exactly where it belongs. A philosophy that cannot survive office paperwork is decoration.

The second boundary is identity.

In his device-authentication essay, anonymous activity belongs first to a local device identity. When login becomes useful, those device identities aggregate under the user account. The account gathers continuity without erasing the local history; the device gets to begin before it has a full user behind it.

Ownership before membership sounds abstract until a database row does it. Let use begin. Let local history form. Let identity widen when the wider identity adds value. That is a humane product sequence because the user starts with contact instead of ceremony.

The third boundary is privacy.

The encrypted public journal is the roughest and most revealing specimen. Sarthak stores encrypted entries where the public internet can see the blobs, while the key remains with the person. He names the method's security limits plainly, which makes the product instinct easier to see. The praise here goes to the relation: public route, private meaning, no stored secret, no account dependency.

Privacy appears as a property of construction. The object gives different readers different affordances. A stranger can fetch the page and still fail to read the life inside it. The person with the key can open the same public object and recover the private one.

That is close to the Markov/Homebase split from the personal side. Markov asks what a system may safely expose from above. Homebase asks what a person chooses to project from inside. Sarthak's journal asks the smaller version of the same question: what can be public without becoming readable by the wrong reader?

The fourth boundary is agency.

His applied-AI essay walks through LLM calls, structured outputs, chat history, retrieval, tool use, routing, parallelization, evaluator loops, and orchestrated workers. The strongest part is the restraint. Start from requirements. Watch cost. Keep the workflow understandable. Give agents runtime discretion where the product earns it.

That is the taste agentic engineering needs. The prompt is a small part of the system. The real work sits in the context boundary, the tool surface, the memory policy, the cost shape, the evaluation loop, and the stop condition. Sarthak's writing has the early, practical version of that taste. He is interested in AI because it helps products act, and he is cautious because action without boundaries becomes mess.

The fifth boundary is craft.

His AI essay comes from use, not posture. He uses models heavily and still keeps the builder in the loop: read the docs, understand the requirement, fix the bug, gather feedback, choose the product, care about the result. That is amplification in the only register that matters, inside actual work. The model can write a large share of the code. The builder is still the one holding the product boundary.

This is why the smaller human essays matter. Luckiest man is a gratitude inventory. Finding Ambition asks why someone would ever give a founder's vision the hours a founder wants. Those pieces keep the machine work inside a life made of bodies, parents, friends, tiredness, worship, pride, doubt, and incentives. Personal AI that forgets that life becomes a capable system pointed at a missing subject.

Even the rough data-structures note belongs to the pattern. It looks under stacks, queues, arrays, objects, roots, heads, pointers, hash buckets, and termination conditions until the named abstraction becomes an operation in memory. That habit is upstream of boundary design. A person can decide what should cross a boundary only after seeing what the boundary is made of.

The project archive gives the philosophy a body. Printscreen makes screenshots presentable while emphasizing privacy. Mindmaps turns thought into shareable structure. ByteMason points at agentic full-stack app creation. Lumeon includes quota management, cited knowledge search, and multimodal AI. QuantDash includes an embedded AI tutor. The list is uneven in the way a real builder's list should be uneven. It is a workshop with fingerprints still on the bench.

So here is the homage.

Sarthak furthers Hari and Markov Blanket philosophy by building small artifacts in the same direction. Shared vocabulary is optional. The work keeps asking the question the graph cares about: what does this object let through, what does it hold back, who gets to read which layer, and when does the boundary change?

That question shows up in a PDF, an auth table, a journal, a screenshot tool, a mind-map product, an AI tutor, a cited knowledge search, an agentic app builder, and a personal website. The range matters because philosophy spreads through ordinary objects before it spreads through agreement. A builder makes the abstraction cheap enough to touch.

Hari can name the principle. Markov Blanket can make it a product. Sarthak's site gives the principle another proof: boundary-first software is already being rediscovered by builders who are trying to make useful things.

That is the nicest thing a philosophy can learn about itself. Someone else is building toward it without needing the name.