tolicodes✳

AN OPEN NOTEBOOK / TOLI ZASLAVSKIY

Decisions, detours,
and what I learned.

A home for stories from my engineering work. Some are about things I built. Others are about changing my mind, asking for help, or getting people to build something together.

28 of 28 stories

DropboxInfluence

A good tool still needs an owner

The hard part of shared Storybook tooling was finding an ownership model teams could support.

Read the storyClose story

At Dropbox, teams were rebuilding components because they could not easily discover or try what other teams had built. There were many disconnected Storybooks, but hosting, build integration and ownership were unresolved.

I started a working group, talked with teams about their constraints, and helped establish a shared standard with Bazel and Chromatic. An important turning point was understanding that the platform team was concerned about ongoing maintenance. I built support around sustaining the tool, alongside the technical groundwork.

My accounts describe adoption across more than 20 teams. The lesson was that installing a useful tool and making it viable across an organization are different kinds of work.

Related readingCurrent résumé
Link to this story
Jimini HealthArchitecture

Reusable without becoming a framework

Support a real partner’s workflow while keeping the genuinely shared pieces reusable.

Read the storyClose story

At Jimini, I owned the Task Center implementation and designed a model for a partner integration. The challenge was to support the workflow we actually had while leaving room for other partners without overabstracting.

I separated shared capabilities, such as document uploading, from task-specific metadata. Tenant adapters handled differences in operations instead of forcing every partner into identical behavior. A partner that needed current appointment information from its provider could use a different implementation of the same operation.

The boundary was practical: share the concepts that were already common, and keep variations explicit.

Related readingCurrent résumé
Link to this story
PlayBuddyProduct judgment

The feature I loved wasn’t the job users had

I built for my own power-user habits. Users wanted to decide what to do tonight.

Read the storyClose story

I spent roughly one to two months building a Tinder-like event-discovery interface for PlayBuddy. I was a power user, going to around five events a week, and thought swiping would be a great way to find the next one. Other people were not using it.

At first I changed the experience and explained how it worked, assuming the problem was understanding. When I finally talked with users, they described a different job: “What do I want to do tonight?” or “What do I want to do this weekend?” They wanted a quick decision, not half an hour of swiping.

I gave up that use case because it did not justify navigation-bar space. I focused on the date bar and filters so people could narrow to their free day and the experience they wanted.

Related readingCurrent résumé
Link to this story
Jimini HealthReliability

A reminder depends on what is true now

Appointment updates and cancellations made notification delivery a reconciliation problem.

Read the storyClose story

I revamped a notification pipeline spanning partner ingestion, scheduling, queues, push delivery and in-app messages. An appointment could change or disappear after its reminders had been scheduled; other changes could affect whether a reminder was still relevant.

We listened for events, checked the provider as the source of truth, and reconciled what needed to be created, updated or deleted. An event was a reason to check the current state, rather than an instruction to send another message.

The delivery path also crossed Android, iOS and in-app messaging, with different foreground behavior. We used a broad test suite to explore edge cases.

Related readingCurrent résumé
Link to this story
HOVERInfluence

Making time for infrastructure

Better testing tools helped. Engineers also needed time to use them.

Read the storyClose story

At HOVER, tooling and documentation helped unit tests grow, but not as quickly as we wanted. Feature deadlines still competed with infrastructure work.

A colleague and I worked on making dedicated infrastructure time a team habit. He helped make the case to management; I supplied tooling, documentation and technical groundwork. Infra Day later became a recurring infrastructure sprint, supporting changes such as TypeScript, React Hooks and GraphQL.

My published HOVER account reports roughly 85% unit-test coverage after three Infra Days. The useful part of the intervention was making space for engineers to address the work together.

Related readingMy HOVER story
Link to this story
DropboxProduct judgment

Give people a reason to download the app

A desktop-client recommendation made more sense during a difficult browser upload.

Read the storyClose story

Our Dropbox team wanted more people to adopt the desktop client. Instead of repeating a generic download message, we looked for moments when it would solve an immediate problem. Bulk browser uploads could take a long time, encounter errors, or be interrupted when someone closed the tab.

I implemented the upload trigger in the campaign system, created and ran campaigns, configured rollout, and prepared charts and a dashboard. I also worked on validation and proposed when the prompts should appear. The message connected the desktop client to the upload problem the person was already experiencing.

My recollection is that the contextual approach increased desktop downloads, and desktop users were more likely to become paid customers.

Related readingCurrent résumé
Link to this story
DropboxProduct judgment

Turn a mockup into a testable release

Make implementation costs visible, then work backward from the hypothesis.

Read the storyClose story

As a project DRI at Dropbox, I would go through a design with the PM and make the cost of each feature explicit. An entire mockup was not automatically the minimum useful release.

I presented a shopping list of implementation costs and worked backward from the hypothesis we needed to test. Then I prepared the architecture, initial implementations and tasks so other engineers could execute.

This made the choice between scope and delivery easier to discuss with product partners.

Related readingCurrent résumé
Link to this story
DropboxArchitecture

Modernizing without losing the download journey

Improve a high-traffic install surface while preserving its core job.

Read the storyClose story

I led a six-month effort to modernize Dropbox’s /install page using TypeScript, React and Python. It was an important surface for moving people into the desktop product.

The work had to improve maintainability while preserving the download journey. My résumé and project accounts describe fewer recurring bugs and less maintenance afterward.

The central constraint was that a cleaner implementation still had to do the existing product job reliably.

Related readingCurrent résumé
Link to this story
DropboxReliability

Help the right fix happen

Clarifying a fragmented system mattered even when a colleague wrote the final solution.

Read the storyClose story

A desktop-client detection issue crossed several download surfaces and had moved between on-call engineers. I mapped the redundant flows, gathered engineers with context, and helped prioritize a durable fix.

I worked with colleagues on investigation and followed up on an edge case after validation. A colleague arrived at the final solution.

This is an example of clarifying a shared problem and enabling another engineer to finish the fix. My contribution was not authorship of the final patch.

Related readingCurrent résumé
Link to this story
DropboxReliability

Deleting code is still a delivery project

An onboarding cleanup crossed more pathways and teams than I first expected.

Read the storyClose story

I owned removal of a legacy onboarding quiz under a deadline. The scope was larger than expected and touched important new-user pathways.

I broke the work into incremental changes, coordinated with partner teams, delegated and paired on implementation, and tested the major pathways. My project account describes six phases of pull requests.

I delivered within the deadline despite the initial underestimate. Making dependencies and risk visible was part of delivering the change.

Related readingCurrent résumé
Link to this story
DropboxProduct judgment

Look next door before building

A neighboring team already had an implementation worth understanding.

Read the storyClose story

For a Dropbox onboarding project, I investigated with another team whether its existing implementation could meet our needs.

Working with them gave us a substantially simpler approach. The important decision happened before committing our own engineering time: learn how the neighboring solution worked and what compatibility work remained.

Related readingCurrent résumé
Link to this story
DropboxInfluence

An answer in one person’s memory doesn’t scale

Shared knowledge needs discovery, ownership and a practical way to use it.

Read the storyClose story

I worked on making internal answers easier to discover instead of repeatedly resolving the same questions in Slack. My project and résumé accounts describe renewed use of an internal question-and-answer tool and work with knowledge-management stakeholders.

I treated discovery as an infrastructure problem. A useful answer that remains in one person’s memory or an isolated conversation is difficult for the next engineer to find.

Related readingCurrent résumé
Link to this story
DropboxReliability

Make operational signals usable

Dashboards and alerts should help the on-call engineer understand what is happening.

Read the storyClose story

I improved on-call dashboards, metrics and alerts so the team could understand service health and diagnose problems more consistently.

The work included consolidating operational reporting. I wanted the signals and the reporting workflow to be easier for the person doing the work to use.

Related readingCurrent résumé
Link to this story
HOVERArchitecture

Start with why people avoid writing tests

PickleJS grew out of repetitive setup and tests that were hard to read.

Read the storyClose story

At HOVER, asking engineers to write integration tests was not enough. Tests required repetitive setup and became difficult to read and maintain. I tried existing approaches, wrote tests, and used that work to identify repeated patterns.

I built PickleJS on the Cucumber approach, using reusable English-language steps and separating selectors from scenarios. I paired with teams on their actual cases and kept the modules extensible.

The framework was released as open source. Its design came from the reasons people were avoiding the work.

Link to this story
HOVERArchitecture

Make the standard development path easy

A service template brought tests, deployment and environments together.

Read the storyClose story

As HOVER’s frontend organization grew, I worked on containerizing frontend applications and improving local development.

My published account describes a boilerplate that let a team add a microservice with CI/CD, tests and three environments through a pull request. The goal was to make the standard path easy to use as more teams and services appeared.

Related readingMy HOVER story
Link to this story
HOVERReliability

Visual testing beyond components

A 3D house renderer needed checks that ordinary UI assertions could not provide.

Read the storyClose story

I worked with HOVER’s computer vision team on visual regression checks for its 3D house renderer.

This extended testing beyond isolated frontend components into a product whose output was visual and harder to judge with ordinary assertions. My published account and résumé establish that collaboration.

Related readingMy HOVER story
Link to this story
Jimini HealthProduct judgment

A specialized product still has switching costs

People bring habits and accumulated context to the tools they already use.

Read the storyClose story

While working on a specialized companion product, I encountered an adoption problem: people already had history and habits in an existing general-purpose tool. I found that moving my own accumulated context was difficult too.

The lesson I draw is that specialization alone does not remove switching costs. I would want to understand those habits early and how a new product fits into the existing workflow.

Related readingCurrent résumé
Link to this story
Costume Lab and independent projectsAI engineering

Build the evaluation tool alongside the product

A result could look good while missing the idea, inventing a detail, or taking too long.

Read the storyClose story

I built Costume Lab to turn an idea into a costume assembled from real products. A visually appealing result could still miss the idea or invent a product detail.

I built evaluation tooling alongside the product to retain runs, compare a candidate with a pinned baseline, inspect separate quality dimensions and track total workflow time. The tool preserves the code and settings behind each result.

The point was to make the tradeoff visible so a change could be examined rather than judged from one attractive image.

Related readingPortfolio
Link to this story
Independent projects and Jimini HealthAI engineering

Keep agent work reviewable

Independent approaches still need explicit integration decisions and checks.

Read the storyClose story

I use AI coding agents on separate components and competing approaches, with independent checking before integration. I also use shared architecture notes, reusable review instructions, Git worktrees and isolated environments.

The work remains a series of decisions about which changes to keep, how to combine them, and what evidence is needed before delivery. Generated code volume alone does not tell me whether a result is useful.

Related readingCurrent résumé
Link to this story
MastercardReliability

Testing a payments experience

Coverage mattered as a way to make changes safer to examine.

Read the storyClose story

At Mastercard, I worked on Click to Pay features using Vue, React and Node.js. My résumé describes visual regression and unit/integration testing with Playwright and Jest.

I worked on extending the test coverage around the product. Company transaction volume gives context for the domain; it is not a measure of the impact of my individual changes.

Related readingCurrent résumé
Link to this story
Novel CommerceArchitecture

Building and coordinating a startup launch

Smart contracts were one part of bringing a product to launch.

Read the storyClose story

I built Solidity smart contracts for Novel’s NFT platform and helped launch the product. My résumé also describes hiring and guiding contractors and contributing to technical and product decisions involving Shopify and Apple Wallet.

The story combines implementation with coordination: a working contract was part of a broader product and team effort.

Related readingCurrent résumé
Link to this story
DropboxLearning

Ask for the working session sooner

More individual effort wasn’t solving a missing-context problem.

Read the storyClose story

I spent about three weeks trying to understand an internal platform through scattered questions. Eventually I booked a focused session with an engineer who knew it. In roughly an hour, they explained the system and where to look.

I had treated more individual effort as the only way forward. What I needed was a bounded working session with someone who had the missing context.

Related readingCurrent résumé
Link to this story
InfraGen and technical writingLearning

When useful tooling grows beyond its purpose

InfraGen kept acquiring ideas and refactors until the scope became the problem.

Read the storyClose story

In my Technical Principles writing, I describe starting InfraGen with a clear aim and repeatedly following new ideas and refactors until the project grew beyond its original purpose.

I also describe large pull requests that were difficult for other people to review. The lesson I wrote down was to define a useful minimum, solve smaller problems and keep changes reviewable.

Related readingTechnical Principles
Link to this story
Writing and engineering teamsInfluence

Make knowledge usable by the next person

Pairing, documentation and shared practice made technical knowledge easier to reuse.

Read the storyClose story

I have paired with developers, written reusable tooling and documentation, presented to groups and helped others prepare technical talks.

At HOVER, that included sitting with teams as they adopted PickleJS and extending it around their real scenarios. I later wrote the Frontend Infra Book after finding gaps in frontend system-design preparation, and organized NYC LeetCode Squad for technical interview practice.

Link to this story
Smartly.io interview preparationLearning

Learn the domain by writing the book

Preparing for Smartly showed me how much I had underestimated advertising.

Read the storyClose story

I was interviewing for a Technical Account Manager role at Smartly.io, coming from software engineering with very little Facebook marketing knowledge. Initially I thought advertising was mainly about making an attractive ad and linking it to a website. Research showed me how much I had underestimated it.

I read documentation and articles, watched tutorials and courses, and spoke with people who worked in marketing. I organized what I learned into FBAdsBook, covering the advertising workflow, tools, optimization and why a product like Smartly could help.

Writing gave me a concrete way to explain the domain I was learning. I later joined Smartly.

Related readingFBAdsBook
Link to this story
Frontend Infra BookLearning

Turn preparation into a shared resource

Frontend system-design preparation had a gap I wanted to help fill.

Read the storyClose story

While interviewing for architectural frontend roles, I found that most system-design preparation focused on backend or distributed systems. That left a gap for the work I was preparing to discuss.

I pulled together knowledge from my career, asked other engineers for help, and did mock interviews. I organized the material into the Frontend Infra Book and released it as an open-source guide so other frontend engineers could use it and contribute.

Related readingFrontend Infra Book
Link to this story
NYC LeetCode SquadInfluence

Make practice a shared habit

I wanted a more social way to prepare, so I started NYC LeetCode Squad.

Read the storyClose story

I had spent years working as a frontend engineer without doing much LeetCode. When the job market changed, I recognized that this limited my options. I wanted a social way to practice and stay accountable.

When I could not find the right group, I started NYC LeetCode Squad. A community member offered temporary space, and the first event brought together 26 people, from students to experienced engineers. Members began teaching one another, including helping beginners with the fundamentals.

We organized small practice pods where senior members could explain their reasoning while helping junior members.

Related readingNYC LeetCode Squad
Link to this story
Smartly.io interviewLearning

Stop performing a stereotype of the role

Nerves made me imitate how I thought a salesperson should sound.

Read the storyClose story

During a Smartly interview for a customer-facing role, I was nervous about coming from engineering without much sales experience. I assumed a salesperson needed to sound very confident and forceful. Under pressure I overcompensated and felt I never established a connection.

Afterward, I reflected on the assumption I had brought into the conversation and what I actually wanted to learn in the role. The useful lesson was to explain my real experience and curiosity rather than perform a stereotype.

Related readingCurrent résumé
Link to this story

Kept here for the long run.

This is a growing collection of edited personal accounts, current as of November 2026. Figures are recollections or attributed historical reports, rather than independently audited results. Each story keeps the limits of what I can substantiate.

The public edition lives in versioned source and is available as ordinary files. Save a copy, read it later, or use the structured version to explore it another way.