Svetlana PichuginaDownload CV

Open to Senior / Staff AI Engineer roles · Remote · Tbilisi

Svetlana Pichugina

Senior AI Engineer · Agentic systems & AI infrastructure

AI agents that make it to production and stay there.

10+ years in software. At Smartcat I build agentic systems and the infrastructure under them: an AI testing system that runs 4–7-hour nightly regressions 80–90% hands-off, reusable workflow components on Temporal, and an AI agent platform adopted by 20+ enterprise companies.

Get in touchDownload CV

Impact

Smartcat · 2024–2026
  • 80–90%of 4–7-hour nightly regression runs completed with no one watching
  • 4 weeksto ship a product estimated at 4 months, as a team of 4 engineers
  • 22 of 37reusable agentic workflow components built by me, shared by 2 pipelines
  • 20+enterprise companies on an AI agent platform I co-architected
  • 29AI-generated merge requests in one day; ~90% passed tests on the first try
  • 10%of company revenue from an AI localization platform, 6 months after launch
Selected systems

Production AI, not demos.

01 · Smartcat · built solo in 2 months

Autonomous testing infrastructure

An AI-powered system that runs the nightly end-to-end regression across 7 file types on its own. When infrastructure breaks, deterministic recovery with 11 allowlisted actions fixes about 80% of failures before an LLM is ever called.

  • Temporal
  • Playwright MCP
  • Self-healing workflows
  • Deterministic recovery
80–90%hands-off nightly runs
~80%infra failures fixed without an LLM
TeamAdopted by 2 engineering teams; mentored 3 QA engineers through architecture sessions and hands-on pipeline work
02 · Smartcat · Temporal

Reusable agentic workflow components

Wrote 22 of the 37 building blocks behind two engineering pipelines. Because they compose, automated merge request testing plugged in about a week after we decided to build it.

~1 weekto add automated MR testing
03 · Smartcat · team of 4

AI-assisted delivery

Up to 12 AI-assisted tasks in flight at once, with specs and automated tests deciding what merges. A product estimated at four months shipped in four weeks.

4 mo → 4 wkestimate vs delivery
04 · Smartcat · 20+ enterprise companies

AI agent platform

Co-architected and launched a production agent platform: agentic workflows over RAG and vector retrieval with Weaviate, context management and tool calling. Launched in six months.

20+enterprise companies
05 · Smartcat · ~7 enterprise customers

SCORM editing agent

Retrieves the relevant course content, proposes edits as diffs for a person to approve, then applies only what was approved.

1–2 h → 15 minper editing task
06 · Smartcat · internal platform

MCP servers for engineering

Slack, meeting transcripts, YouTrack, GitLab and Playwright checks over MCP, so context survives from discussion to merged code.

100+developers daily
07 · Smartcat · reliability

AI incident response

One place to investigate incidents: Langfuse traces, Kubernetes state, GitLab changes, tickets, Slack threads and logs.

−70%investigation time
Frontend for AI products

An agent is only as good as the screen where people check its work.

Most of my ten years went into frontend: Vue, React, Nuxt, Next. At Smartcat I lead the frontend architecture of our AI products too.

Smartcat · AI products12×

faster frontend releases

Microfrontends, a shared design system and a UI component library. Deploys went from 2 hours to 10 minutes.

Smartcat · AI localization4 months

to ship an enterprise AI product

Took part in its architecture and built the frontend foundation. Adopted by Amazon and other enterprise customers, 10% of company revenue within six months.

N1.RU · Cian Group15×

faster pages on a real-estate marketplace

Moved key apps from Vue 2 to Nuxt with SSR, added Apple Pay and Google Pay to checkout, built Mapbox property search.

Interface
  • Vue 3
  • Nuxt
  • React
  • Next.js
  • TypeScript
  • Pinia
  • TanStack Query
  • Vite
  • Vitest
  • Storybook
  • Playwright
  • SSR
  • Microfrontends
  • Design systems
Across the team

Frontend standards for several AI products: architecture decisions, code and design reviews, mentoring engineers.

My own product · live · you are on its domain

I built an AI assistant for the job hunt.

One place from a vacancy to an offer. It reads a posting, finds the gaps against a study plan, looks up how the company hires, drafts the letter and gets you ready for each interview stage. I use it in my own search.

  • Solo: product, design, code, ops
  • In production from day one
  • Built with Claude Code
Open the appFree sign-in with Google.
Job posting · RemoteSenior AI Engineer
  1. Fit check
  2. How they hire
  3. Cover letter
  4. Note to HR
  5. Interviews
  6. Outcome
Next stepDrafting your cover letterWritten from your resume and the company research, then checked for AI-sounding phrases.
Brush up before the interview
Asked often · EvalsAsked often · RAGAsked less · LangGraphWorth growing · Fine-tuning

  1. Vacancy → study plan

    Paste a posting. The model maps each requirement onto the study guide and ranks the gaps: asked often, asked less, worth growing.

  2. Search by resume

    Finds open roles that fit the resume and checks every link, so closed postings never reach the list.

  3. Research and letters

    How the company hires, from public reviews. A cover letter checked for AI-sounding phrases, and a note to someone on the team.

  4. Every interview stage

    Before a stage, a forecast of what they will ask. After it, a debrief from your notes that sets up the next one.

Under the hood
  • Astro SSR
  • React islands
  • PostgreSQL + Drizzle
  • OpenRouter (Claude), Zod-validated output
  • Background jobs
  • Blue-green deploys
  • Docker + Caddy
  • Vitest
  • PostHog
How I work

Four habits behind the numbers

  1. Agents propose, people decide

    My agents show every change as a diff to approve before applying it. That is what made enterprise customers trust them.

  2. Deterministic first, LLM second

    An allowlist of 11 recovery actions handles ~80% of failures in my testing system. The model only gets the cases code cannot settle, and Langfuse traces show what it did.

  3. Context is the product

    Most agent failures are missing context. MCP and retrieval come first, clever prompts second.

  4. Ship in months, then grow

    Four weeks against a four-month estimate, a testing system in two months. Scope is cut, quality is not.

Toolbox
  • TypeScript
  • Node.js
  • OpenAI API
  • Anthropic Claude API
  • MCP
  • Temporal
  • Playwright MCP
  • LangGraph
  • Weaviate
  • Langfuse
  • PostgreSQL
  • Kubernetes
  • Python
Experience

From enterprise web to agents

Ten years across fintech, compliance, real estate and B2B platforms before AI became the day job.

  1. 2024 — now

    Senior AI Engineer

    Smartcat

    Autonomous testing infrastructure, 22 reusable workflow components on Temporal, an agent platform for 20+ enterprise companies, AI localization. Technical lead across 7–15 engineers.

  2. 2021 — 2024

    Senior Full-Stack Engineer

    Trustform

    Configurable KYC and compliance platform: 2× faster reviews, 4× faster customer customizations. TypeScript, React, Fastify.

  3. 2020 — 2021

    Frontend Engineer

    N1.RU · Cian Group

    Up to 15× faster pages after moving to Nuxt SSR; Apple Pay and Google Pay in checkout.

  4. 2018 — 2020

    Frontend Engineer

    LC Group

    Fintech MVP for Vietnam in 8 months that paid back its cost within 4 months of launch.

  5. 2017 — 2018

    Frontend Developer

    IT Megastar

    Enterprise features in React, Angular and Vue.

  6. 2015 — 2017

    Software Engineer

    Tensor · SBIS

    Business platform for 4.5M+ customers; 25% fewer production incidents.

What colleagues say
“Sveta is one of the strongest engineers I’ve had the opportunity to work with. She combines exceptional execution speed with consistently high quality, often delivering in days what might take others weeks.”
Svetlana Ivanova
“I particularly valued her ability to turn complex business requirements into practical, well-designed technical solutions. She combines strong engineering expertise with a genuine understanding of product goals.”
Kseniya KhortBusiness Analyst

Hiring for AI in production? Let’s talk.

Open to Senior and Staff AI Engineer roles, fully remote. Email is the fastest way to reach me.