expert_search ยท diff
git:20260902.af2a47c to git:20260906.f9cef19
53 added, 7 removed. Audit A to A.
---
name: expert_search
- description: A specialized matchmaker designed to connect startups with the most qualified members of the community for due diligence, evaluation, investment, or mentorship. It generates a ranked, evidence-based list of experts from the `person_profile` dataset based on a deep analysis of a startup's industry, technology stack, and business model.
+ description: Rank SICTIC members for a startup's due-diligence needs using stored investor profiles and the startup profile. Use for domain, functional and strategic expertise matching.
---
- The output Insight's `config_cache_key()` covers the expert-search objective, both
- shared ranking config sections, and runtime target, exclusion, and `top_k`
- options. The ranking engine constrains permitted profile IDs with JSON Schemas,
- supplies those schemas to LiteLLM, repairs the response, and validates it
- locally before rendering Markdown.
+ # Expert search
+ Find members whose professional experience fits a startup's due-diligence needs.
+
+ ## Inputs and outputs
+
+ The async `expert_search(startup_name, target_experts=None,
+ exclude_experts=None, top_k=16)` returns one Markdown ranking in
+ `list[InsightFile]`. Optional person filters accept names, emails or LinkedIn
+ IDs using the [shared ranking selection rules](../ranking/SKILL.md).
+
+ Save through `InsightFile` in the startup's insights directory as
+ `expert-search-<startup>-<model>.md`.
+
+ ## Workflow and dependencies
+
+ Resolve the dataset through `ensure_startup_dataset`. Preserve a manual ranking
+ override before preparing dependencies. Otherwise request the normal
+ `startup_profile`, insert its content into the configured expert-search objective,
+ and rank stored `investor_profile` insights from `sictic-members`.
+ The objective prioritizes direct domain experience, then functional expertise
+ and strategic experience.
+
+ Direct calls require an existing member roster and investor profiles; they do
+ not discover or refresh members. Bulk refresh registers `startup-profile` and
+ `investor-profile` prerequisites; their execution follows registry scope.
+
+ Use `InsightFile.find(selection="reusable")` after reading the startup profile.
+ Freshness depends on indexed revisions of the startup and `sictic-members`,
+ plus the objective, shared ranking/structured-output configuration, supplied
+ startup-profile content and runtime filters/limit. A changed startup profile
+ invalidates the ranking even without reindexing. Edits to stored investor
+ profiles or the roster alone are not tracked by this cache.
+
+ ## Side effects and failure behavior
+
+ Startup preparation may import or index documents and generate its profile.
+ Ranking reads member profiles, calls the shared ranking engine when needed,
+ and saves the completed report. It performs no outreach.
+ Input and generation failures propagate without saving a new ranking.
+ Missing indexed revisions prevent verified cache reuse.
+
## Usage
```bash
- conda run -n sictic-env python -m skills.harness /expert_search "<STARTUP_NAME>"
+ conda run -n sictic-env python -m skills.harness '/expert_search "<STARTUP_NAME>"'
```
+
+ The direct CLI accepts `--startup`, comma-separated `--include` and `--exclude`,
+ and `--top-k` (default 16). The harness accepts only the startup argument.
+
+ ## References
+
+ - [Implementation](expert_search.py)
+ - [Objective](../../config/expert_search/objective.md)
+ - [Shared ranking contracts](../ranking/SKILL.md)
+ - [Insight freshness](../standards_and_architecture/SKILL.md#selection-and-freshness)