Levi

AI Engineer · Founding Engineer at RoleCall · Former Mobile Team Lead

Eight years of iOS, including leading a mobile team of up to 15 people, and now building AI systems full-time: agent orchestration, evaluation and fine-tuning. Strong background in math and computer science. I think in architectures and abstractions, and I like bringing organic order into chaos: non-disruptive solutions, automation, and making hard concepts easy to understand. Based in Paphos, Cyprus.

Experience

Founding Engineer

—RoleCall Studios

Dec 2025 — Present

RoleCall is an AI roleplay platform: an agentic narrator, a story director, persistent world state and a creator toolkit, with open signups and paid plans since June 2026. I designed and built the original agent stack, then developed it together with co-founders Chi and Nemo in a pnpm/Turborepo monorepo (about 1,300 of my commits since April 2026).

  • Original architecture: the agentic turn loop and tool-calling harness, the chat UI with SSE streaming, the story director that plans arcs and runs a shadow world, the immersion trackers that turn prose into game state, the memory pipeline around Chi's Compendium, and inference routing with end-to-end encrypted bring-your-own-key.
  • Visual Novel mode: took an early runtime to production. Story graph canvas, graph-aware prompting, an inline stage-command protocol, scene modes, endings, relationship meters with rubrics, and an in-editor test mode (about 105k lines of TypeScript including tests).
  • Character Minds: per-character episodic memory, mood, drives and theory of mind, on a pgvector semantic layer that also powers discovery search and lorebook retrieval.
  • Billing and paid launch, shared with Chi: pricing model, weekly token pools charged per generation, a request-metered entry tier, Creem integration and the closed-beta exit.
  • PlotPoints: moved my community RP arena into the product mid-round; round two closed at 1,943 votes from 482 voters.
  • Also: voiceover with per-character voices, the Dramaturg read-only story assistant, scene-search tools for every in-app agent, send-latency work (5.3 s to 2.5 s before generation starts), and incident write-ups.
TypeScriptNext.jsReactSupabasePostgreSQLpgvectorTurborepoSSEPython

Mobile Team Lead

—Croc Code

May 2021 — Apr 2022

Led a mobile team of up to 15 people: hiring, performance and planning, Agile practices, and client scoping. Set up CI/CD across all mobile projects (Fastlane, Firebase, GitLab, App Center, TestFlight) and owned project architecture.

SwiftMVVMCleanSwiftVIPERFastlaneGitLab CIJira

Senior iOS Engineer

—CROC

Aug 2017 — Apr 2022

iOS development from engineer (2017) to senior engineer (2018): SDKs for ID-document recognition and liveness checks, BLE peripherals, a video social network with conferencing, and architecture and estimation for client projects.

SwiftObjective-CC++AVFoundationCoreBluetoothCoreDataFirebaseXamarin

Software Developer

—Geomatix Laboratory

May 2014 — Aug 2018

macOS and iOS development: maintained the YoYo social app, and refactored the NertOut keyboard app, including a migration from SQL to Core Data without data loss.

Objective-CSwiftCore DatamacOS

Lecturer — Math Logic & Competitive Programming

—Moscow Power Engineering Institute (MPEI)

2020 — 2022

Taught math logic and co-organized competitive programming electives from scratch, from curriculum and materials to branding and student trips.

AlgorithmsData StructuresMathematical Logic

Selected projects

A roleplay benchmark for LLMs: 27 scoring dimensions, adversarial multi-turn runs with a user simulator, and a public blind-vote arena that showed LLM judges disagreeing with real users. Open dataset on HuggingFace with 2,800+ downloads.

QLoRA fine-tunes of a persona model on a single 16 GB GPU, from dataset building to GGUF export and serving with tool calling.

RAG extension for SillyTavern: lead contributor (136 of 189 commits). Also two smaller SillyTavern extensions.

Skills

AI & LLM engineering

Agent orchestrationTool callingPrompt architectureLLM evaluationLLM-as-judge calibrationQLoRA fine-tuningRAG & embeddingsInference routing

Web & Infra

TypeScriptNext.jsReactSupabasePostgreSQLpgvectorNode.jsPythonRailwaySSE streaming

iOS & Mobile

SwiftObjective-CMVVMVIPERCleanSwiftCore DataFastlane

Research & Teaching

Cognitive modellingPhilosophy of mathCurriculum designTechnical writing

Education

Postgraduate studies (PhD track) — Theory of Informatics

Moscow Power Engineering Institute (MPEI)

2018 — 2022. Research in applied mathematics and informatics; taught math logic and competitive programming.

M.Sc. — Applied Mathematics

Moscow Power Engineering Institute (MPEI)

2016 — 2018

B.Sc. — Mathematics and Computer Science

Moscow Power Engineering Institute (MPEI)

2012 — 2016

Languages

Russian (native)English (professional)German (elementary)Polish (elementary)

Get in touch

Interested in working together or just want to talk AI? Reach out at levi@l3vi4th4n.ai