Savva Lazarev

Product Architect | Belgrade, Serbia

Summary

Product architect with hands-on operational experience across the full mediabuying pipeline. Progressed from project management through mediabuying leadership to technical lead — architecting a system for end-to-end creative production management. Focus: removing friction from processes, making data flow naturally, and enabling teams to focus on what they're good at.


Experience

Concepta LLC (Mar 2025 -- present)

Performance marketing company. Parent brand aggregator with several hundred million in annual revenue across multiple brands.

Technical Lead (~4 months, current)

  • Architected and led development of an internal creative production platform (see product description)
  • Led a team of developers: requirements definition, architecture decisions, code reviews, feedback loops
  • Managed implementation: user acceptance, change management, addressing team concerns about new workflows
  • Built programmatic analytics for ad performance and team productivity
  • Priority: 100% platform adoption — system designed so data flows naturally without extra manual actions from users

Head of Mediabuying (~4 months)

  • Led mediabuying operations and campaign performance
  • Deep involvement in financial planning and budget allocation
  • Identified key performance-driving angles through systematic analysis of ad data and creative patterns
  • Interviewed and onboarded new team members, expanding the mediabuying team
  • Continued managing operations (contractor payments, KPI tracking) — PM role was never backfilled

Project Manager (~4 months)

  • Financial operations: contractor payments, salary verification, KPI calculations for the team
  • Identified systematic friction in workflows (manual handoffs, fragmented tools, scattered data)
  • Optimized own processes to free up time for higher-impact work
  • Participated in hiring: interviewing candidates for operations and mediabuying roles

Projects

Transcription + RAG Knowledge System (architect, ongoing)

  • Designed and architected a call transcription pipeline: 862+ calls processed over 9 months
  • Pipeline: audio → speech-to-text (speaker diarization) → AI summarization → structured storage + Telegram delivery
  • RAG service for semantic search over call history: FastAPI + Pinecone (vector DB) + PostgreSQL (metadata), 3-stage search pipeline (metadata filtration → semantic search → post-processing)
  • Conversational agent with iterative query refinement for instant retrieval of decisions, context, and commitments from months of calls
  • Platform-agnostic recording ingestion: workflow designed to pull audio from any device or source into a unified processing pipeline
  • Built to make institutional knowledge instantly searchable; continuing development as a standalone tool

Automated TikTok Content Pipeline (architect, completed)

  • Designed a fully automated posting pipeline for TikTok: content sourcing → editing → scheduling → publishing
  • Device-agnostic architecture: workflow pulls recordings from any source and processes them through a unified pipeline
  • Built for high scalability — designed to run across multiple accounts with minimal manual intervention
  • Analyzed posting performance and iterated on content strategy based on engagement data

Key Skills

Leadership & Operations

  • Process design: identifying bottlenecks, designing systems that eliminate unnecessary manual work
  • Change management: feedback loops, user acceptance testing, managing adoption resistance
  • Financial operations: payments, salary tracking, KPI calculation, budget planning
  • Cross-functional communication: bridging creative teams, mediabuyers, and developers

Technical

  • System architecture: database-heavy design for maximum data capture and analytics
  • AI-assisted development: focus on planning, prioritization, and architecture
  • Familiar with: TypeScript, Python, SQL, Next.js, React, Supabase, PostgreSQL, FastAPI, Pinecone, LangGraph
  • Integrations: Telegram Bot API, Google APIs, AI services (speech-to-text, image/video generation, LLMs)
  • Infrastructure: Railway, Vercel, GitHub Actions, Docker, n8n

Approach

  • Find the bottleneck, fix only that (Theory of Constraints)
  • Evaluate every systematic manual task against its total impact to the system
  • Manual first, automate second — validate before building