Abelardo Carlos Martínez Lorenzo

1. Sep 2024 – Present

   Washington, DC / Rome

   Lead AI Scientist

   The World Bank · ImpactAI

   • Lead an AI science team of six direct reports; own architecture, core backend implementation and code review for a conversational evidence-synthesis platform supporting meta-analysis and development policy.
   • Build Python/FastAPI REST APIs for PDF parsing, information extraction, RAG, entity linking, SQL analysis and summarization, with streaming, failure recovery and automated tests.
   • Design human-in-the-loop evaluation, LLM judges and retrieval benchmarks; improve prompts and models through annotator feedback, user sessions and error analysis.

   + 5 moreShow less

   • Optimize inference cost and latency through domain-specific fine-tuning, encoder models, batching and caching; build Vertex AI training pipelines and synthetic-data workflows.
   • Develop LLM-based taxonomy generation and refinement with human validation to structure and retrieve research evidence.
   • Guide and co-author research on predicting causal effects from natural-language queries, with a 72,000+ query benchmark and a structured LLM-to-encoder pipeline.
   • Own GCP infrastructure, resource planning and cloud budgets; automate GitHub-to-GCP CI/CD with Docker, Cloud Build and Cloud Run.
   • Build terabyte-scale news pipelines for food-crisis research with Pandas, PyArrow and Parquet.

   Came out of thisImpactAI ↗Query2Effect →

2. Jan 2024 – Sep 2024

   Washington, DC / Rome

   Senior AI Scientist

   The World Bank

   • Designed and implemented a two-stage entity-linking system: an e5 retriever followed by a fine-tuned DeBERTa encoder that maps research text to an impact-evaluation taxonomy.
   • Built FastAPI/Docker NLP services; managed GCP resources, container orchestration, API deployment and latency optimization.
   • Built and evaluated LLM prompting pipelines for information extraction and semantic parsing with GPT-4o, LLaMA and Gemini.

   Came out of thisImpactAI ↗

3. 2023 – 2024

   Rome, Italy

   NLP Researcher

   Sapienza NLP · Sapienza University of Rome

   • Designed compact graph representations and encoder-decoder parsers in PyTorch Lightning, cutting training time by 82% and inference time by 78% versus SPRING while preserving parsing quality.
   • Developed multilingual semantic-parsing datasets through structured annotation and GPT-3.5 fine-tuning, evaluated on 1,100 human-annotated sentences across 11 languages.
   • Led research with PhD students on multilingual language understanding and LLM-based parsing, alignment and tagging.

   + 1 moreShow less

   • Teaching assistant for MSc NLP courses; mentored students in research methods and implementation.

   Came out of thisCLAP →MSL →MOSAICo →

4. 2020 – 2023

   Rome, Italy

   NLP Researcher

   Babelscape

   • Designed cross-lingual meaning representations and BART/T5 parsing and generation models; delivered Dockerized inference demos.
   • Built alignment-aware training with cross-attention supervision; co-developed graph-informed adapters and self-distillation for encoder-decoder models.
   • Developed autoregressive ensembling and graph validation, cutting invalid outputs to 0.3% at equal SMATCH on AMR 3.0.

   + 1 moreShow less

   • Co-developed a multilingual corpus with hundreds of millions of silver annotations across four language-understanding tasks and five languages.

   Came out of thisBMR formalism →BMR 1.0 →Cross-lingual aligner →LeakDistill →AMRs Assemble! →

5. Sep 2019 – Jun 2020

   Berlin, Germany

   Research Assistant

   Technische Universität Berlin

   • Designed and led the implementation of a neuro-symbolic framework for knowledge-graph completion, combining embedding models with ontology-based reasoning.
   • Built interchangeable model and OWL2 reasoner interfaces; evaluated TransE, ComplEx and DistMult on DBpedia and LUBM, improving DistMult MRR from 0.28 to 0.35 on LUBM.

   Came out of thisPoderoso →2022 preprint →

6. 2017 – 2018

   Málaga, Spain

   Software Engineer

   Freelance

   • Built a real-time, multiplatform emergency-coordination app with chat, peer-to-peer WebRTC voice, maps, incidents and team assignment.
   • Developed the backend with Node.js, LoopBack and MongoDB, and the frontend with AngularJS.

Spoken languages: English (fluent) · Spanish (native) · Italian (fluent) · Portuguese (intermediate)