---
title: "Extracting a KRT from a manuscript"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Extracting a KRT from a manuscript}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
```

```{r setup}
library(krt)
```

`krt` can extract candidate resources from a manuscript with a deterministic
regex engine (the default, fully offline) or an optional LLM.

## Scan text for identifiers

```{r}
txt <- "We stained with anti-TH (RRID:AB_390204) from Millipore (Cat# AB152)
        and analyzed images in Fiji (RRID:SCR_002285). Data: GEO GSE12345."
scan_identifiers(txt)
```

## Extract a table

```{r}
res <- extract_krt(txt)
as.data.frame(res$krt)[, c("resource_type", "display_name", "rrid")]
```

The result is a normalized, validated `krt_tbl` plus a validation report; every
extraction is provenance-stamped with the engine used.

## Read structured documents

`read_input_text()` handles PDF, JATS/NISO XML, DOCX, and plain text, and
`detect_existing_krt()` parses a Key Resources Table already present in the
document.

```{r}
jats <- system.file("extdata", "examples", "sample.jats.xml", package = "krt")
res <- extract_krt(jats)
res$existing_krt
```

## LLM extraction (optional)

The LLM engine is opt-in and requires a provider and API key. It is
non-deterministic, so it is never the default, and its output is funneled
through the same normalize and validate steps the regex engine uses before you
see a candidate table. (Direct importers such as `import_krt()` only read and
structure the file; normalize and validate them yourself when you need to.)

```{r, eval = FALSE}
cfg <- krt_llm("openai", model = "gpt-4o-mini")   # reads OPENAI_API_KEY
res <- extract_krt("path/to/manuscript.pdf", engine = "llm", llm = cfg)
```

You can register a custom or local provider:

```{r, eval = FALSE}
register_llm_provider("local", function(prompt, llm) {
  # call your local server, return the model's text output
})
extract_krt(txt, engine = "llm",
            llm = structure(list(provider = "local"), class = "krt_llm"))
```