Applied research and development
Test technologies before recommending them
The Lab is PartITech’s applied research space for artificial intelligence in web products. We test architectures, libraries and evaluation methods to distinguish a promising demonstration from a genuinely operable solution.
This work informs our monitoring, open-source tools and architectural choices. It does not replace the analysis specific to each client project.
- Experiment
- Isolate a hypothesis and test it against realistic cases
- Measure
- Compare quality, latency, cost and constraints
- Share
- Document results, including limitations
Areas of work
Topics we put to the test
RAG and search
Chunking, indexing, hybrid search, citations, access rights and answer evaluation.
Models and inference
Models accessed via API or deployed on controlled infrastructure, quantisation, latency and cost.
Documents and data
Extraction, structuring, metadata and preparation of reliable corpora for AI processing.
Product integration
Tooling, observability, security, user experience and integration with existing applications.
Experiment protocol
An experiment must be reviewable
We define the question, dataset, metrics and environment before comparing results. This discipline limits conclusions drawn from hand-picked examples.
Read the publications- 01State a testable hypothesis
- 02Build reference cases
- 03Trace versions and parameters
- 04Measure gaps and failures
- 05Document usage limits
From Lab to product
Turn a finding into an architecture decision
An experiment becomes useful when it informs a trade-off: selecting a model, defining a RAG strategy, sizing infrastructure, establishing an evaluation protocol or abandoning a fragile path.