About

PortMem is built by one person who has been inside this problem for seven years. The first two hires at close: a founding engineer and an enterprise GTM lead.

Andre Bacellar
Andre Bacellar
Founder · CEO

For seven years I worked inside enterprise knowledge-base infrastructure at a major US provider. I watched the same failure mode again and again: the system kept retrieving the deprecated 2017 policy over the active 2024 one, generating confident answers from stale sources, and offering no way for compliance to audit what had been used.

The standard answer was "add more LLM". More agentic loops, longer context windows, fancier rerankers. None of it solved the actual problem, which was that retrieval was being optimized for similarity when regulated buyers needed retrieval optimized for what is currently in force.

I wrote four arXiv papers to fix this from first principles. PhaseGraph calibrates vector and graph scores. BridgeRAG handles multi-hop. RegimeRouter picks the right strategy per query. CAR formalizes the controlling-authority problem and proves it is a different mathematical object from semantic relevance. Those four papers are the engine. PortMem is the product built on them: the system of record for AI work product.

Background

Eleven years in enterprise sales

Before becoming a researcher, I sold complex software into regulated buyers. I have sat in the procurement meeting where a wrong citation kills the deal. I know what compliance officers read for and what they flag.

Seven years in enterprise knowledge infra

I built ingestion, retrieval, and ranking systems at scale for a major US enterprise provider. The patterns that broke in production are the patterns PortMem is designed to prevent.

Four sole-authored arXiv papers

Each one fixes a specific failure mode in retrieval over regulated content. Datasets, scoring code, and reproducibility memos are public. See the research.

Columbia MBA · BS Electrical Engineering

Trained on both sides of the table. Engineering rigor on the product, business literacy on the GTM.

Why now, why me

Three things converged in the past 18 months. Frontier LLMs stopped hallucinating when handed a single relevant passage, so the bottleneck moved from generation to retrieval. The EU AI Act made citation accuracy a legal liability, with high-risk obligations from December 2027. And the research to solve regulated retrieval became tractable.

I sat at the intersection of all three: an operator who had felt the pain, a researcher who could write the papers, and a salesperson who knows what regulated buyers will pay for. PortMem exists because the same person did all three.

Want to talk?