GLEB ROSEV

Early-stage US startup building named AI agent teams that take engineering work from proposal to merged GitHub pull request under human-approved plans.

• Tenant isolation: tenant-scoped query guards and sandboxed code execution.
• Connector backfills made self-healing: stuck-cursor recovery, progress tracking and hard caps for long syncs of large workspaces.
• Code-review agent reliability: fixed ticket-key collisions between YAGNI and external trackers that sent the reviewer to the wrong issue.
• Agent benchmarks: realistic engineering task suites that show whether a prompt, model or harness change makes agents better or worse.
• Risk radar: flags stalled customer runs before the customer reports them.
• Self-improving loop (experimental): a meta-agent reads failed runs, isolates the cause and drafts a fix for the team to review.
• Monorepo build chain reduced to one command for fresh checkouts, worktrees and CI.
• Technical point of contact for pilot customers: demos, debugging live runs, turning feedback into engineering work.

AI receptionist platform for healthcare: voice and text agents for scheduling, care coordination and clinic operations. Sully.ai serves 400+ healthcare organizations.

• Led core engineering for the MENA launch: UAE infrastructure, Azure Blob Storage support, regional configuration and production readiness.
• Removed ~15 s of dead air at pickup on outbound calls: the agent was greeting the caller while the line was still ringing.
• Made agent dispatch reliable on LiveKit: retry-and-verify, no more duplicate agents in one call, no more silent calls.
• Designed UAE audio-residency guardrails so TTS and STT fallbacks stay within approved regional providers instead of silently routing elsewhere.
• Built patient-data validation for outbound campaigns: local phone formats, customer-specific schemas, clinic rules.
• Built high-volume stress testing with headless LiveKit participants, Loki and Mimir validation, and runtime health checks; found and fixed the memory cascade behind burst failures.

AI-powered multi-tenant B2B SaaS platform for wholesale.

• Led a cross-functional team of up to 12 people while staying active in architecture and code.
• Delivered multi-tenant catalog/order workflows with ERP/CRM syncs and feature-flagged trunk-based releases.
• Built an LLM feedback pipeline converting Sales/CS input into product insights (-80% analysis time).

Product engineering across fintech, crypto, AI & healthcare.

• Led a 5-person team across web and mobile product delivery in a shared monorepo.
• Integrated native cryptographic modules into a flagship application with 100,000+ downloads.
• Hardened CI/CD build pipelines and production runtime stability.

Sports tournament platform, full-stack cloud architectures & technical mentorship.

• Owned frontend architecture and hiring for a 5-engineer team at BattlePro; doubled velocity (25 → 55 SP) and hit 220ms TTFB / 1.7s TTI.
• Architected a full-stack virtual networking platform at ITechArt that subsequently attracted $10M+ in investment.
• Delivered 150+ advanced React/JS lectures at TeachMeSkills; curriculum adopted by 1,500+ students/year.