Skip to content

Independent work

Labs

Three projects exploring how people learn a language, follow research sources, and inspect a model’s behavior.

01 / Language learning

Local prototype

Causerie

Read material you choose, keep the phrases you want to learn, and bring them into writing you can save and revise.

Read the Causerie case study
Reading and expression

Reader note

Ce projet demande un peu plus de temps.

Keep the passage with the note.

Revised expression

Il nous faudrait un peu plus de temps pour terminer le projet.

Return to the draft and try the phrase in context.

Workflow illustration with original sample text.

02 / Research tools

Desktop prototype

CiteLoom

A desktop research canvas that keeps search plans, paper candidates, and source notes in one workspace.

Research canvas
  1. 01Plan the search
  2. 02Retrieve and rerank
  3. 03Inspect sources
Architecture sketch of the retrieval workflow.

The retrieval design pairs a BGE reranker with reciprocal rank fusion. An MCP-compatible interface supports inspection and repeatable checks.

03 / Model observability

Local MVP

NeuralLens

A local prototype for inspecting layer summaries and comparing text-model runs after a temporary parameter patch.

Inference comparison

Same prompt

A Baseline runB Temporary patch

Compare outputs and layer summaries

Workflow diagram for a local A/B experiment.

Run a deterministic mock or a local Hugging Face causal language model. Save an experiment as JSON and reopen it for review.