Summary
Software/AI engineer who ships end to end — from durable, event-driven LLM pipelines and a multimodal eval framework to the AWS/Terraform infrastructure and zero-downtime deploys that run them. Helped take an AI video-generation platform from MVP to public launch across several codebases (Python, C#, TypeScript), owning REST APIs, a card + crypto payments system, serverless GPU workers, and the CI/CD behind it.
Skills
Languages: Python, C#, TypeScript, SQL, Bash, HCL (Terraform)
Backend & APIs: FastAPI, Pydantic v2, AWS Lambda Powertools, REST API design, JSON-Schema structured output, JWT / API-key auth
Async & Orchestration: Inngest durable step functions (webhook suspend/resume, timeout handling), n8n, AWS EventBridge Scheduler
Cloud & Infra: AWS (Lambda, S3, API Gateway, VPC, IAM, EventBridge), Terraform multi-env modules, Docker, RunPod serverless GPU, self-hosted VPS with blue-green deploys, nginx, systemd, Tailscale/WireGuard, GitHub Actions
Databases: PostgreSQL (RDS), psycopg2, stored procedures, connection pooling
AI / ML: LLM gateways (OpenRouter/Portkey), structured output, promptfoo evals & LLM-as-judge, text-to-image / image-to-video providers, LoRA fine-tunes, audio source separation, LaMa/OpenCV inpainting
Payments & Web3: Stripe, crypto checkout, Solana/Helius on-chain verification
Game Development: Unity (URP, WebGL), VR/AR (XR Toolkit, MRTK3, Oculus SDK), Photon, DOTween
Sandwatch — Software / AI Engineer
Berlin, Germany (Remote) · Feb 2025 – present
Shipped an AI video-generation platform from MVP to public launch, working across Python backend, C# generation engine, cloud infrastructure, TypeScript bots, and GPU workers.
- Built a durable, event-driven AI generation pipeline — a ~20-step Inngest flow spanning script, voice/TTS, character, and multi-stage visual generation that suspends on provider webhooks and resumes with request-ID matching, with strict JSON-Schema structured output at every LLM step.
- Built the quality layer: a promptfoo eval framework with a multimodal LLM-as-judge that grades rendered videos, codified into 11 judge modules inside an automated suite of ~198 tests gated in CI.
- Shipped 30+ REST endpoints on AWS Lambda (Pydantic v2, PostgreSQL) covering agents, presets, characters, generations, and a filterable central library.
- Built the billing/credit system end to end — atomic charge-on-create with source-aware refunds via stored procedures, integrated card (Stripe) and crypto checkout with on-chain verification; reconciled, dispute-free billing.
- Turned single-provider video generation into a multi-provider fallback chain, cutting the failure rate by more than half and reducing average generation time.
- Migrated the voice catalog to a new TTS/cloning provider and scaled it to a 600-voice multilingual library across 12 languages.
- Built two serverless GPU workers: ensemble vocal removal (audio source separation) and automated video inpainting / object removal (OpenCV + LaMa).
- Sole developer of the original node-based C# engine turning a topic into finished long-form video — LLM story generation, text-to-image/image-to-video models, LoRA character consistency, subtitles, cloud assembly.
- Owned deploys end to end: zero-downtime blue-green releases behind nginx, Terraform-managed prod/dev AWS environments, and a Tailscale/WireGuard fix that took deploys from intermittently failing to consistently reliable.
- Built a TypeScript Discord bot with S3 presigned delivery and Solana payment verification, plus a Cloudflare-backed CLI that let non-technical staff publish landing-page content with no deploy.