Evgeny Bazarov
AI systems · Paris
I’m an AI engineer and technical leader in Paris, currently building agent systems for accounting at Allia.
I work across retrieval, evaluation, and production infrastructure. I like the work after the demo, where software meets real workflows and engineering quality starts to matter.
Selected work
- 01Allia
Agent systems for accounting that hold up in production.
Multi-agent copilot, evaluation, and durable ingestion at 340,000 workflows a week.
- 02Gino Legal Tech + Tomorro
Contract tools built with legal teams.
Extraction, clause review, Word redlining, and GraphRAG. Processing time fell 48% at Gino. Satisfaction rose from 75% to 92% at Tomorro.
- 03Besedo
Seven years of production ML for Trust & Safety.
Led up to eight engineers and industrialized 130+ multilingual NLP, vision, and similarity models.
Thinking about
A cited answer can still be wrong. Grounding is a system property. Retrieval, context, tool results, and the final answer all need to agree.
Scale exposes product problems. Production volume reveals unclear workflows and weak feedback loops, not only infrastructure limits.
Human review is architecture. In accuracy-sensitive work, the right escalation and approval path is a product capability, not a fallback.