T Tudor Andrei Dumitrascu

AI engineering for teams shipping serious workflows

Production AI systems from messy source material to measurable product value.

I help companies build AI products, document intelligence workflows, and ML infrastructure that can survive real users, real latency, and real operational constraints.

Start a project View proof
AI Engineer CTO Bucharest, Romania Available for selected consulting

What this site is for

Make technical buyers understand where I create leverage.

This first pass is the product surface: clear routes into services, proof, operating style, and contact. The final copy can get sharper once the structure feels right.

Services

Work packages that map to business outcomes.

01

AI Product Engineering

From product idea to usable AI workflow: prototypes, evaluation loops, APIs, and the surrounding application surface.
  • LLM apps
  • RAG
  • Evaluation
  • FastAPI

02

Document Intelligence

Systems that parse complex documents, extract rules, and turn unstructured material into reliable product behavior.
  • Legal docs
  • Extraction
  • Q&A
  • Pipelines

03

ML Infrastructure

Inference, observability, and data pipelines built for throughput, cost control, and production ownership.
  • Triton
  • Docker
  • Prometheus
  • RabbitMQ

04

Full-Stack Delivery

Backend services and practical product interfaces for teams that need the complete path from model to user.
  • Next.js
  • NestJS
  • PostgreSQL
  • Redis

Proof

Selected work translated into buyer-readable signals.

1,000+ articles/sec

High-throughput News Intelligence

NVIDIA Triton inference pipeline, distributed scraping, extraction, and monitoring for large-scale media processing.

Rule extraction

Legal Document AI

Playbook ingestion, template parsing, and document-aware question answering for legal workflows.

Multi-tenant

Compliance SaaS Platform

KYC/AML workflow platform with graph-style data modeling, API integrations, and production observability.

Acquired project

Computer Vision Tracking

Real-time boat detection and tracking project commercialized through acquisition by a maritime autonomy company.

Operating Model

A practical path from unclear AI ambition to shipped capability.

The sales flow should make it obvious that the work is not just model selection. It is scoping, data reality, application design, production constraints, and feedback loops.

01

Scope the workflow

Define the decision, document, or user journey the system must improve.

02

Build the slice

Ship a narrow version with real data, measurable outputs, and clear failure cases.

03

Harden the system

Add evaluation, latency control, monitoring, deployment, and maintainable APIs.

Now

CTO, Pythia Sociodynamical Technologies

2025

AI Engineer, CogniSync

2022

Co-Founder & CTO, Pythia Sociodynamical Technologies

2022

M.Sc. Artificial Intelligence, University of Bucharest

DSPyPyTorchNLPComputer VisionFastAPIReactPostgreSQLAWSAzureDockerNix

Contact

Bring a workflow, a bottleneck, or a product idea.

Email Tudor GitHub LinkedIn