Senior full-stack engineer:from data model to revenueto productionto monitoringto the user

I own features end to end — data model, asynchronous backend, production UI, infrastructure, monitoring. Production AI & DevEx is where I go deepest: multi-agent delivery pipelines, custom MCP servers, model cascades.

What I've shipped

1. 01AI assistant for lead processingmonthly channel revenue tripled from ~$47K to ~$141K; response time slashed from ~8 min to under 30 s with a 10× drop in API costs.−
   problem

   managers took about 8 minutes to reply and only during shifts — hot leads went cold, and chat-manager headcount was expensive.

   constraints

   Load is spiky and dialogs are live: operators need the queue updated in real time, not on page refresh.

   what I built

   The autonomous assistant replies in under 30 seconds at any hour: high-intent leads receive immediate qualification. Designed a confidence-based routing cascade in FastAPI and Redis: lightweight models handle routine qualification and escalate to frontier models (Claude / GPT-4) only on ambiguous inputs, slashing API bills 10×. Integrated real-time CRM Tool Calling and a live WebSocket operator queue where specialists monitor dialogs seamlessly. After a 3-month controlled trial, the channel moved fully to AI.

   trade-off

   A cascade requires custom routing instead of one catch-all model. The payoff is a tenfold reduction in API bills.

   stack

   PythonFastAPIReactRedisWebSocketLLM APIsClaudeOpenAI

   Read the write-up (in Russian)Get in touch >
2. 02Credit reports & scoring on 20M rowsthe product turned profitable in its first week: net profit +40%, EPC doubled from ~$1.9 to ~$3.8.+
   problem

   launching a credit-intelligence product under strict SLAs with scoring computed against historical datasets.

   constraints

   Credit and scoring data: strict isolation, sub-second report retrieval by phone number.

   what I built

   Led a 3-engineer team: owned technical design, API contract, and credit bureau integration for decisioning. Designed data schemas for applications, reports, and scoring; built an asynchronous FastAPI backend querying a 20M-row scoring dataset via PostgreSQL composite indexing and Redis caching.

   trade-off

   Composite indexing and Redis caching took extra upfront query profiling, but enabled sub-second scoring over 20M rows.

   stack

   PythonFastAPIPostgreSQLRedisReact

   Read the write-up (in Russian)Get in touch >
3. 03Kafka event-driven push enginestreaming clickstream under peak load, delivery <500 ms latency, with zero notification loss.+
   problem

   at peak loads, service push notifications and analyst telemetry clickstream were getting lost.

   constraints

   what I built

   Engineered an Apache Kafka event-driven streaming pipeline connecting clickstream telemetry directly into an asynchronous FastAPI push engine and live React UI: producers emit events, consumers batch and process them sub-second without delays. Monitored with OpenTelemetry and Grafana alerts; end-to-end covered with Playwright.

   trade-off

   Kafka added cluster operations, but eliminated telemetry drops and guaranteed instant trigger delivery.

   stack

   PythonFastAPIApache KafkaCeleryDockerKubernetesOpenTelemetryGrafana

   Read the write-up (in Russian)Get in touch >
4. 04Autonomous multi-agent delivery pipelinefeature turnaround cut from ~2 h to ~40 minutes; with 100% contract test coverage and zero regression leaks.+
   problem

   routine feature modules and regression testing consumed hours of engineering time as the codebase expanded.

   constraints

   Strict context isolation between subagents and automated rejection of broken invariants.

   what I built

   Designed an isolated multi-agent engineering delivery pipeline: requirements decompose into strict Pydantic DTO contracts, tests precede implementation in a strict TDD red phase, subagents execute in isolated contexts, and an automated reviewer gate enforces architectural invariants via custom MCP servers.

   trade-off

   Formalizing contracts takes 10 minutes upfront, but saves hours of integration debugging.

   stack

   PythonFastAPIClaude CodeMCPPydanticVitestPlaywright

   Get in touch >
5. 05Task-and-rewards product & anti-fraudscaled to 1,000 daily active users; in 3 months, achieving 100% payout auditability and eliminating fraudulent submissions.+
   problem

   scaling paid user tasks required fraud protection against duplicates and forged completion proofs.

   constraints

   Hundreds of completions daily: blocking fraudulent submissions without hurting legitimate user payouts.

   what I built

   Modeled the full task lifecycle from claim to payout: responsive React dashboard for task discovery and proof upload with real-time WebSockets, automated chatbot re-engagement flows, an asynchronous moderation queue with anti-fraud rules, and deep UTM/referrer attribution tracking paired with cohort activity metrics.

   stack

   ReactTypeScriptRedux ToolkitPythonFastAPIPostgreSQLWebSockets

   Get in touch >
6. 06Campaign canvas & 100K-row tablemarketing revenue with zero engineering intervention — ~$80K/month; scoring table load time slashed from 8 s to under 1 s.+
   problem

   marketing depended on engineering to launch push campaigns, while the 100K-row scoring table took 8 seconds to load.

   constraints

   Heavy scoring queries in real time and enabling autonomous marketing operations without production risk.

   what I built

   Engineered a visual no-code campaign canvas allowing marketers to configure audience segments and delivery triggers independently without developer tickets. Optimized PostgreSQL queries with composite indexing, offloaded heavy tasks to Celery/Redis, and implemented client-side row virtualization: the UI smoothly renders 100K scoring rows with instant sub-second response.

   trade-off

   Virtualization requires fixed row dimensions, but keeps the DOM lightweight regardless of table size.

   stack

   PythonFastAPIReactPostgreSQLRedisCelery

   Read the write-up (in Russian)Get in touch >

1. 01 · 2022 — 2024

   Appbooster

   remote · Aug 2022 — Aug 2024 · 2 yrs

   Mobile app marketing and user-growth platform: promoting mobile applications across 26 countries.

   
   role

   Full-Stack Engineer

   product

   Autonomous LLM lead-processing agent, model cascade, task platform & microservices

   achievements

   • Tripled monthly channel revenue from ~$47K to ~$141K and slashed lead response time from ~8 min to under 30 s: led development of autonomous LLM lead agent with real-time CRM Tool Calling (3-month controlled trial)

   • Designed confidence-based model routing cascade in FastAPI and Redis: cut LLM API cost 10× per lead with zero quality drop, escalating to Claude / GPT-4 only on ambiguous inputs

   • Achieved 100% payout auditability and scaled task platform to 1,000 DAU in 3 months: async moderation queue with anti-fraud rules, real-time React WebSockets dashboard, and cohort UTM/referrer attribution

   • Decomposed legacy PHP monolith into async Python/FastAPI microservices: boosted throughput by 40% with zero downtime, built React CRM webhook hub, and lifted SPA conversion ~1.5× (Lighthouse 58 to 87)

   PythonFastAPIReactTypeScriptRedux ToolkitPostgreSQLRedisApache KafkaAWSDockerWebSocketsLLM APIsPlaywrightJestAnt Design
2. 02 · 2024 — now

   ID Finance

   remote · Sep 2024 — present · 2 yrs

   Digital lending and credit intelligence platform. Promoted to Senior for cross-functional impact and technical leadership.

   
   role

   Senior Full-Stack Engineer (05/2025–Pres.) · Full-Stack (09/2024–05/2025)

   product

   AI document parser, credit scoring on 20M rows, autonomous 5-agent pipeline, Kafka push engine, campaign canvas

   achievements

   • Automated borrower document verification in under 15 s and cut manual review load by ~60%: hierarchical AI document parser (AWS S3, Vision LLMs, PostgreSQL pgvector + BM25 hybrid search, reranking) via LangGraph multi-agent workflow, MCP, and Pydantic schemas

   • Led 3-engineer team: credit-report product profitable in week one (+40% net profit, EPC $1.9 to $3.8); async FastAPI backend scoring over 20M-row dataset via PostgreSQL composite indexes and Redis

   • Cut feature turnaround from ~2 h to ~40 min with 100% contract test coverage via autonomous 5-agent delivery pipeline (PRD, TDD, FastAPI/React, reviewer gate) on Anthropic/OpenAI Python SDKs, Pydantic contracts, and custom MCP servers

   • Event-driven pipeline on Apache Kafka (Consumer Groups, DLQ) on AWS & Kubernetes: streaming clickstream into async FastAPI push engine — sub-second delivery (<500 ms) and zero loss under peak load

   • Generated ~$80K/month in marketing revenue with zero engineering intervention and cut 100K-row table load from 8 s to under 1 s: visual campaign canvas (React, FastAPI, Celery/Redis) and PostgreSQL query optimization with row virtualization

   PythonFastAPIasyncioLangGraphPostgreSQLRedisApache KafkaCeleryReactAWSDockerKubernetesGitLab CI/CDOpenTelemetryClaude CodeMCPAnthropic/OpenAI Python SDK