mădălin tatarciuc

ai engineer • applied research & systems builder

i've spent the last few years fine-tuning llms at scale, building production ml systems, training computer vision models, and shipping ai to real users. now i'm focused on getting intelligence on the hardware, building systems that run on-device, at the edge, where latency matters and the cloud isn't always there. also working on rl agents for autonomous uav control.

founder at superbird labs — we're building embedded intelligence infrastructure.

experience

jun 2024 — present

founder — superbird labs

  • building embedded intelligence infrastructure
  • developing ai systems for edge deployment and on-device intelligence
  • researching agent architectures, long-term memory systems, and small-model pipelines
jun 2025 — dec 2025

engineering team lead — salestools

  • architected ai-driven sales platform processing 10m+ monthly events
  • designed llm-based reply generation trained on 48m+ emails and 36m+ linkedin messages
  • built and fine-tuned conversational agents under real-time response constraints
  • optimized inference latency and cost across cloud and bare-metal gpu deployments
jun 2024 — jun 2025

co-founder & engineer — dotworlds

  • built generative media pipeline (text → image → video)
  • designed prompt optimization and orchestration logic
  • scaled to 1,000+ users within one month of launch
oct 2024 — may 2025

ai automation engineer — plant an app / bazinga ai

  • built llm-powered automation workflows and rag systems
  • designed chunking, embedding, retrieval, and evaluation strategies
apr 2023 — apr 2024

business dev partner & romania chapter lead — omdena ai

  • led romanian ai community
  • coordinated open-source ml projects including cv systems for deforestation detection
earlier

skillab • gameloft • coreto.io

research & projects

kalkan defense

research and engineering of reinforcement learning agents for uav control, policy learning, and autonomous decision-making

hugging face

published ner models; trained and evaluated rl agents with custom environments

superbird app

personal intelligent unit (piu): an ai companion with multimodal input, long-term memory, and adaptive behavior

education & other

buildspace

ai builder cohort (2024)

ase bucharest

informatics economics (2019–2020)

certifications

deep learning, machine learning, computer vision

kendo

national team athlete, 1 dan

contact

madalintat@gmail.com

+40 754 461 974

bucharest, romania