recombinase 0.1.3

Template-guided document synthesis: extract structured content from source pptx files and recombine into a canonical template

recombinase

Biology: a recombinase is an enzyme that extracts DNA fragments and recombines them into new molecules using a homologous template as the structural guide. This package does the same for PowerPoint documents.

Template-guided pptx synthesis. Take a styled "filled example" slide in a .pptx/.pptm template, a folder of per-record YAML data files, and produce a populated output deck — one slide per record, visually identical to the template because the fill operation duplicates the source slide and replaces text in-place by shape name.

Install

pip install --user recombinase

Dependencies: python-pptx, pyyaml. That's it.

Windows 11 install notes

On a standard Windows 11 machine with Python installed:

python -m pip install --user recombinase

The --user flag installs into your user profile (%APPDATA%\Python\Python3XX\site-packages) which doesn't require admin rights and usually slips past managed-device policy that blocks system-wide installs. After install, the recombinase command is on your PATH at %APPDATA%\Python\Python3XX\Scripts\. If recombinase --version isn't found after install, use python -m recombinase.cli --version as a fallback.

From source (dev)

git clone https://github.com/terry-li-hm/recombinase.git
cd recombinase
pip install --user -e ".[dev]"

Concepts

Three steps, loosely coupled by file:

1. Template — a .pptx/.pptm file with at least one slide where every field you want to populate is a named shape (e.g. a text box named Consultant_Name).
2. Config — a small YAML file declaring which shape name on the template corresponds to which data field. One config per template. Templates change; configs go with them.
3. Data — a directory of per-record YAML files, one file per record (e.g. one per consultant). Each file has a flat map of field names to values. List values become bullet paragraphs automatically.

The template config is intentionally per-template rather than hardcoded in the package. Same package, different config → different template. You can build a CV pack, a use-case slide deck, or a client case study collection with the same tool and three different configs.

Usage

1. Inspect a template

Discover the shape names on each slide — structural metadata only, never the actual text content. Safe to share the output.

recombinase inspect "path/to/template.pptm"

Example output:

File: /path/to/template.pptm
Slide count: 1

=== Slide 1 (layout: 'Blank') ===
  - 'Consultant_Name' | type=TEXT_BOX (17) | text_chars=12 | paras=1, runs=1
  - 'Role_Title' | type=TEXT_BOX (17) | text_chars=18 | paras=1, runs=1
  - 'Summary_Body' | type=TEXT_BOX (17) | text_chars=140 | paras=2, runs=3
  - 'Background_Bullets' | type=TEXT_BOX (17) | text_chars=220 | paras=5, runs=5

2. Scaffold a config

Generate a starter config file from the template's shape names:

recombinase init "path/to/template.pptm" --output template-config.yaml

This reads the shape names from slide 1 and writes a config like:

template: /path/to/template.pptm
source_slide_index: 1
clear_source_slide: true

placeholders:
  consultant_name: Consultant_Name
  role_title: Role_Title
  summary_body: Summary_Body
  background_bullets: Background_Bullets

Edit the left side (data field names) to match how your records are keyed. For example, if your YAML data files use name: not consultant_name:, rename the left side:

placeholders:
  name: Consultant_Name
  role: Role_Title
  summary: Summary_Body
  background: Background_Bullets

3. Write per-record data files

Create a directory with one YAML file per record. Filenames become the sort order:

cv-data/
├── 01-jane-doe.yaml
├── 02-john-smith.yaml
└── 03-alice-wong.yaml

Each file is a flat map — lists become bullet paragraphs:

id: jane-doe
name: Jane Doe
role: Senior Consultant
summary: >-
  Twelve years across global wealth management with a focus on
  regulatory data and risk modelling.
background:
  - Bank A — Risk modelling lead (2010-2015)
  - Bank B — Head of analytics (2015-2020)
  - Bank C — CDO, Asia Pacific (2020-present)
key_skills:
  - Risk modelling
  - Governance
  - Wealth data architecture

The field names on the left must match the keys in your template config's placeholders: section.

4. Generate the output deck

recombinase generate \
  --config template-config.yaml \
  --data-dir cv-data/ \
  --output output/deck.pptx

Produces a populated pptx with one slide per YAML file. If clear_source_slide: true in the config, the original example slide is removed from the output.

One-line end-to-end

After the config exists:

recombinase generate -c template-config.yaml -d cv-data/ -o out.pptx

Design notes

Why duplicate a filled example slide?

The alternative is creating slides from a layout and writing text into empty placeholders. That approach loses any hand-tweaks the template designer made (custom colours, tweaked positions, decorative shapes, non-placeholder elements). Duplicating a known-good filled slide inherits 100% of its visual styling by design — deepcopy of the shape tree carries every property.

Trade-off: the template must contain one "canonical good example" slide to clone from. This is usually natural for CV templates and pack-prep work.

Rich text and the flattening caveat

When a value is written into a shape with shape.text_frame.text = "...", rich-text runs within that shape (bold name, italic subtitle in one text frame) collapse to the placeholder's default run style. For most modern consulting templates this isn't an issue — each styled fragment lives in its own shape. If your template has a multi-run placeholder, either split it into separate shapes or accept the flattening.

Variable-length lists

List values in the YAML data become separate paragraphs in the target text frame, inheriting the placeholder's paragraph-level bullet formatting automatically. No bullet markers in the source data — the template supplies them. A consultant with three background bullets and another with seven both work without any config change.

Warnings, not errors

If a config references a shape name that doesn't exist, or a record is missing a field, generate produces a warning but continues. This is deliberate: partial output is more useful than total failure during iteration. Pass --strict if you want non-zero exit on warnings.

Recommended file layout (Windows + OneDrive)

For work that involves colleague personal data (CVs, HR records, etc.), keep the package install and the data separate:

%USERPROFILE%\AppData\Roaming\Python\...\   # package install (local, not synced)
C:\Users\<you>\OneDrive - <Org>\Pack\       # work product (synced, backed up)
├── template\
│   ├── CV_template.pptm                   # the canonical pptx
│   └── template-config.yaml               # mapping shape names → data fields
├── cv-data\                                # per-consultant YAML records
│   ├── 01-jane-doe.yaml
│   ├── 02-john-smith.yaml
│   └── 03-alice-wong.yaml
└── output\
    └── deck.pptx                           # generated

The package itself is generic tooling and lives in your Python site-packages — it has no opinions about any particular template or data set. The work data and configs live in OneDrive where they're backed up, versioned by OneDrive's history, and remain inside your organization's sanctioned storage. Run recombinase commands with full paths pointing at OneDrive locations:

cd "C:\Users\<you>\OneDrive - <Org>\Pack"
recombinase inspect "template\CV_template.pptm"
recombinase init "template\CV_template.pptm" -o "template\template-config.yaml"
recombinase generate -c "template\template-config.yaml" -d "cv-data" -o "output\deck.pptx"

Development

git clone https://github.com/terry-li-hm/recombinase.git
cd recombinase
pip install -e ".[dev]"
ruff check src/ tests/
ruff format src/ tests/
mypy src/
pytest

Scope (v0.1)

• ✓ Inspect: print template structural metadata
• ✓ Init: scaffold a config from shape names
• ✓ Generate: populate template from YAML records
• ✗ Extract: reverse direction (pptx → YAML) — v0.2 — needs a sample source file for structure before it can be implemented reliably

License

MIT