Lettria gave AI models the same homework of writing down every fact in a document and connecting them, and tracking output validity and seconds per document
Public source
Publisher name
Public post
We gave AI models the same homework: - here's a document - write down every fact in it - how do those facts connect Then we graded them. We also tracked the unglamorous…
Company
Lettria
Lettria's GraphRAG improves your RAG accuracy on average by 30% while adding transparency to your GenAI Apps
- Industry
- IT Services and IT Consulting
- Location
- Paris, FR
- Company size
- 11–50 employees
About Lettria
Lettria brings business and technology teams together on a no-code collaborative framework, to guide them through every step of a text processing project : Step 1 / Data Collection & Prepration Step 2 / Data Annotation & Ontology Management Step 3 / NLP Engine Customization Step 4 / Demo and output management Get started for free on our website : www.lettria.com Short on time? Ask our support team to help you kickstart your project. 👀 hello@lettria.com 👀
See moreLatest activity
Latest activity from Lettria
5 signals
Products & Services
Lettria's Perseus converts documents into a typed graph in three stages: Analyze, Map, and Emit.
Presence & Recognition
Lettria's team worked with Magnus Helander and The Imago Platform team to define the ontology and pressure test results.
Research & Knowledge
Lettria's team at The Imago Platform® used vector search and a knowledge graph to pull mounting height data from thousands of PDFs and achieve 100% accuracy.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
Collibra
Collibra benchmarked RAG systems and found that the old version outranked the current version on 45.5% of queries.
Research & Knowledge
Inception42
Inception42 reports that the post-graph-rag model achieved 10.4% overall performance on a temporal reasoning question, 10.0% on multi-session reasoning, and 10.5% on knowledge update reasoning compared to 71.2% on Zep/Graphiti and 60.2% on full-context GPT-4o.
Research & Knowledge
LlamaIndex
LlamaIndex benchmarked GPT-6 Astra on hard document parsing and extraction tasks.
Research & Knowledge
Quytech
Quytech published a blog article covering Agentic RAG, its architecture, comparison with traditional RAG, key benefits, business use cases, and implementation considerations.
Research & Knowledge
Towards Data Science