DESIGN PARTNER INTAKEREAL CORPUS WANTEDREV: 0.1
Bring us your messy docs.
BaseMouse is looking for small, real corpora from teams building AI agents, RAG apps, devtools, security/compliance workflows, or internal knowledge assistants. We'll turn your docs and real questions into cited, checksummed context packs and a retrieval-quality baseline.
Redacted docs are fine. We care about realistic structure and query language.
you send: docs + real questions
we return: context-pack examples + eval suite + retrieval baseline
next step: decide whether BaseMouse helps enough to pilotAgent builders
Your agents answer from docs, tools, runbooks, policies, or customer records and need grounded, auditable context.
RAG / LLM app teams
You have retrieval quality questions and need golden-query regression checks before tuning embeddings or graph retrieval.
Governance teams
You need evidence: which documents grounded an answer, which versions were used, and how to audit stale or sensitive context.
WHAT TO SENDLOW FRICTION
One source, one workflow, real questions.
Start small. The goal is to measure whether BaseMouse improves one concrete agent workflow.
- 20–100 docs — markdown, exported pages, PDFs, runbooks, support macros, or redacted samples.
- 10–20 real questions — the prompts users or agents actually ask, not only document titles.
- Source type — Notion, Confluence, Google Drive, Git repo, PDFs, Slack export, or markdown folder.
- Success bar — what a useful context pack must include for your team to trust it.
To: devsupport@basemouse.com
Subject: BaseMouse design partner intake
Company:
Use case:
Source type:
Doc count:
Example questions:
What would make this pilot successful?BaseMouse response:
- retrieval baseline report
- context-pack examples
- golden-query eval suite
- recommendation: pilot / not yet / needs adapterWHAT YOU GETMEASURED, NOT VIBES
A concrete retrieval-quality readout.
We will import the smallest useful corpus, create a golden-query suite, run lexical and hybrid baselines, produce context-pack examples with citations/checksums, and identify whether BaseMouse needs a corpus adapter, learned semantic embeddings, or graph retrieval work before a broader pilot.