3d-Printing on Backend Engineering Strategy Tools

Optics — Can We Print a Lens?

Wed, 03 Jun 2026 00:00:00 +0000

The question: can a 3D printed part function as a lens? Not a perfect optical instrument — but something that focuses or diffuses light usefully.

A Fresnel lens is the natural starting point. Instead of a thick curved lens, a Fresnel collapses the profile into a flat plate with concentric stepped rings — each ring approximating the angle of the equivalent curved surface. Much thinner, printable in flat layers.

Step 1 — Print and test

Two approaches to try:

FDM — clear PLA Printed on the Kobra X. Surface finish from FDM is rough at the layer scale — likely too much scatter for a clean focus, but worth seeing what you actually get.

Resin — clear resins Better surface resolution and the material can be optically transparent after curing and polishing. More promising. Several clear resins ordered for this:

ABS-Like 2.0 Clear — structural, less brittle
Standard Clear — higher detail
High Clear / Transparent resin

See 3D Printing for the full resin inventory.

Not started. Print attempts and results to follow.

Blender Python — Procedural Mesh for 3D Printing

Tue, 02 Jun 2026 00:00:00 +0000

Write a Python script that builds geometry programmatically using Blender’s bpy API, then export as STL. No manual modelling — the script is the source of truth, re-running it regenerates everything.

This pattern is useful when you have a family of similar parts (different sizes, repeated structures, parametric variations) or when the geometry is defined by rules rather than artistic decisions.

When to use this vs. alternatives

	Blender Python	OpenSCAD	CadQuery
Booleans, extrude, modifiers	✓ built-in	✓ CSG only	limited
Sculpt / organic shapes	✓	✗	✗
Parametric constraints	manual	manual	✓ strong
Python ecosystem	✓ full stdlib	✗ own language	✓
Interactive viewport preview	✓	✗	✗
Export to STL	✓ one call	✓	✓

For repetitive mechanical geometry with booleans (holes, sockets, cutouts), Blender Python is the fastest path if you already know Python. The interactive viewport lets you catch geometry problems before exporting.

Core pattern

Every script follows the same structure:

import bpy, bmesh, math, os

# 1. Create a primitive — it becomes bpy.context.object
bpy.ops.mesh.primitive_cylinder_add(vertices=64, radius=5, depth=3)
obj = bpy.context.object

# 2. Edit vertices directly via bmesh
bpy.ops.object.mode_set(mode='EDIT')
bm = bmesh.from_edit_mesh(obj.data)
for v in bm.verts:
 if v.co.z > 0:
 v.co.x *= 0.9 # taper the top
bmesh.update_edit_mesh(obj.data)
bpy.ops.object.mode_set(mode='OBJECT')

# 3. Boolean modifier to cut a hole
bpy.ops.mesh.primitive_cylinder_add(radius=2, depth=3.2)
cutter = bpy.context.object
mod = obj.modifiers.new("Hole", 'BOOLEAN')
mod.object = cutter
mod.operation = 'DIFFERENCE'
bpy.context.view_layer.objects.active = obj
bpy.ops.object.modifier_apply(modifier=mod.name)
bpy.data.objects.remove(cutter, do_unlink=True)

# 4. Export
bpy.ops.object.select_all(action='DESELECT')
obj.select_set(True)
bpy.ops.wm.stl_export(filepath="/tmp/part.stl", export_selected_objects=True)

Build complexity by wrapping each operation in a function, then calling it in a loop over a size or parameter list.

Running the script

Open Blender → switch to the Scripting workspace
Click New or Open to load your .py file
Click Run Script (▶) or press Alt + P

Output and errors appear in the system console (Window → Toggle System Console on Windows, or launch Blender from a terminal on macOS/Linux).

Key API surface

Primitives

bpy.ops.mesh.primitive_cylinder_add(vertices=32, radius=r, depth=h)
bpy.ops.mesh.primitive_cube_add(size=s)
bpy.ops.mesh.primitive_plane_add(size=s)

All primitives land at the world origin and become bpy.context.object.

Transforms

obj.scale = (sx, sy, sz)
obj.location = (x, y, z)
obj.rotation_euler = (rx, ry, rz) # radians
bpy.ops.object.transform_apply(scale=True, location=False, rotation=False)

Apply transforms before any bmesh edits — otherwise vertex coordinates are in local (pre-scale) space, and your edit positions won’t match world coordinates.

bmesh (direct vertex / edge / face editing)

bpy.ops.object.mode_set(mode='EDIT')
bm = bmesh.from_edit_mesh(obj.data)
for v in bm.verts:
 if v.co.z > 0:
 v.co.x *= 0.9
bmesh.update_edit_mesh(obj.data)
bpy.ops.object.mode_set(mode='OBJECT')

Boolean modifiers

mod = target.modifiers.new("Name", 'BOOLEAN')
mod.object = cutter
mod.operation = 'DIFFERENCE' # or UNION or INTERSECT
mod.solver = 'EXACT' # more reliable than FAST for tight geometry
bpy.context.view_layer.objects.active = target
bpy.ops.object.modifier_apply(modifier=mod.name)
bpy.data.objects.remove(cutter, do_unlink=True)

Apply and remove the cutter immediately — leaving stale cutter objects causes confusion on subsequent runs.

Joining objects (single boolean cut for a grid of holes)

Rather than applying one boolean per hole, join all cutters first:

for obj in cyl_objects:
 obj.select_set(True)
bpy.context.view_layer.objects.active = cyl_objects[0]
bpy.ops.object.join()
cutter = bpy.context.active_object
# now do one boolean cut on the plate

One boolean operation is faster and produces cleaner topology than N serial cuts.

STL export (Blender 4.x / 5.x)

bpy.ops.wm.stl_export(filepath="/abs/path/part.stl", export_selected_objects=True)

Collections (organising multi-part output)

col = bpy.data.collections.new("Round Bases")
bpy.context.scene.collection.children.link(col)
for c in obj.users_collection:
 c.objects.unlink(obj)
col.objects.link(obj)

Parametric families

The main loop pattern — build one function that takes dimensions, call it for each size:

SIZES = [10, 15, 20, 25]

for w in SIZES:
 obj = make_clip(width=w)
 obj.name = f"clip_{w}mm"
 bpy.ops.object.select_all(action='DESELECT')
 obj.select_set(True)
 bpy.ops.wm.stl_export(
 filepath=f"/output/clip_{w}mm.stl",
 export_selected_objects=True
 )

Put all tuneable values at the top of the file in a # CONFIG block. This makes the script easy to hand to Claude and say “change the hole diameter to 7mm and add two more rows.”

Clearing the scene before a run

Add this at the top when iterating interactively — otherwise re-running doubles the objects:

bpy.ops.object.select_all(action='SELECT')
bpy.ops.object.delete()
for block in bpy.data.meshes:
 bpy.data.meshes.remove(block)

Viewport layout before export

Spread objects out so you can visually review them before committing to an export:

xoff = 0
for obj, w in zip(objects, widths):
 obj.location.x = xoff
 xoff += w + 5

For print batches, sort largest-first and pack into rows within your bed dimensions (shelf bin-packing against PLATE_W × PLATE_H).

Working with Claude

The config-block pattern pairs well with LLM iteration. Because all tuneable values sit in one place and each operation is a named function, you can describe changes in plain language:

“Change HOLE_DIAM to 6 and add a third text line below line 2” — Claude edits two constants and adds a make_line() call.
“The holes in rows 2 and 3 need 3mm more clearance from the edge” — Claude adjusts x_origin offset computation.
“Add a chamfer around the perimeter of the top face” — Claude adds a bmesh loop and inset operation.

The workflow is: run → look at viewport → describe the problem → apply updated script → repeat. Iteration is fast because the viewport gives immediate visual feedback and the script regenerates from scratch each run.

Limitations

Booleans are fragile on non-manifold geometry. If a cutter face is coplanar with the target, or vertices are nearly coincident, Blender’s solver can produce garbage. Add a small bleed (0.5–1 mm) so cutters fully penetrate surfaces.

No parametric constraints. Unlike CadQuery, there is no “keep this face parallel to that face” system. Dimension changes cascade manually. This is manageable when the config block is the only place numbers live.

Script state accumulates. Re-running in an existing scene doubles the objects. Clear the scene first (see above) or check for existing objects by name before creating.

Garage — Scripted Parts — hole box and other physical builds using this approach
Rack Support Brace — step-by-step: script → renders → headless → CI/CD

Procedural Mesh — Blender Python & AI-Assisted Geometry

Tue, 02 Jun 2026 00:00:00 +0000

Using Python scripts inside Blender to generate, manipulate, and export geometry — rather than modelling by hand. The config block is the model. Changing a dimension means editing a constant and re-running, not touching a mesh.

Two related but distinct approaches:

Parametric generation — the script builds geometry from scratch. Holes, plates, text engravings, export to STL, automated renders. The shape lives entirely in code.
Mesh manipulation — an existing STL is imported and cut up. Different problem: no parametric handle, working with whatever the downloaded mesh gives you.

Both involve AI in the loop — either iterating on the Python with Claude, or using AI to determine cut planes on existing geometry.

Stages

Step		Status
0	Manual Blender modelling — Packat & Klart badge and scout emblem casting master; learning the tool before scripting	Done
1	Approach documented — Python scripted geometry, config-block iteration pattern	Done
2	Rack Support Brace — first implementation; flat plate, boolean hole grid, engraved text, automated renders	Done
3	Headless render — Blender 5.1.2 + Dagger pipeline; STL → multi-angle PNGs without opening Blender; smoke-tested and running in CI	Done
4	Scout Buckle — complex geometry; compound curves, functional tongue mechanism, tolerances that matter	In progress
5	CI/CD pipeline — STL or script change → GitHub Actions → Dagger → PNG artifacts; procedural-mesh-pipeline	Done
6	AI feedback loop — describe correction → updated script → re-run, without opening Blender	Planned
7	Dragon Split — mesh manipulation; cut an existing articulated dragon STL into printable segments with connectors	Not started

Notes

The rack brace (step 2) was easier than expected — straightforward geometry, minimal iteration. That is why the pipeline steps did not get built yet: the manual workflow was fast enough that the overhead of automating it was not justified for one part.

The scout buckle (step 4) is the harder test. Compound curves and functional constraints mean the script has needed partial rebuilds rather than config-block edits. That is the case that justified building the render pipeline properly.

Steps 3 and 5 ended up built together rather than sequentially. The pipeline lives in a separate repo — procedural-mesh-pipeline — Dagger + GitHub Actions, Blender 5.1.2 headless, Cycles CPU. Renders run natively on the CI runner (amd64); locally the pipeline is used for smoke-testing the image only. Blender has no official ARM64 Linux binary so local renders under emulation are impractical.

The dragon split (step 7) is a different class of problem — remixing rather than generating. Worth keeping in the same project because the tooling overlaps (Blender Python, STL export) even if the approach does not.

Rack Support Brace — Programmatic 3D Print

Tue, 02 Jun 2026 00:00:00 +0000

220×150×8mm support brace for the server rack. 40 through-holes in an alternating-spacing grid, B.E.S.T engraved into the top face, exported to STL — all generated by a Python script running in Blender. No manual modelling.

The same script that builds the geometry now also sets up cameras, runs EEVEE renders from four angles, and saves PNGs alongside the STL. The renders above are the direct output.

Downloads

↓rack_support_brace.py ↓rack_support_brace_220x150.stl

The script

Flip EXPORT_STL / EXPORT_RENDER at the top of the file, point EXPORT_DIR at your folder, press Alt+P in the Blender Script Editor.

The config block at the top of the file is the only place dimensions live:

PLATE_X = 220.0 # mm
PLATE_Y = 150.0 # mm
HOLE_DIAM = 7.0 # mm
N_COLS = 10
ROW_SPACING = 20.0 # mm

Iterating means editing a constant and re-running — not touching a model.

See Blender Python for 3D Printing for the general pattern behind this.

Step 1 — Render from script ✓

Done. Script builds geometry → exports STL → places cameras at four angles (top, front, side, iso) → EEVEE renders → PNGs saved to EXPORT_DIR/renders/.

Pipeline next steps (headless render, CI/CD, AI feedback loop) are tracked in the Procedural Mesh project.

The Printed Part

Finished rack support brace installed in the rack.

Resin Badges — Packat & Klart

Tue, 02 Jun 2026 00:00:00 +0000

Resin-printed ‘packat & klart’ badges for the scouts heading out on their first overnight trip, i.e. ‘Hajk’.

We had a meeting where we went through what one should bring, then ran a competition where the kids got points for guessing what was in the pack I brought along. They got a badge if they did well — they all passed. Success.

First attempt at producing physical scout items rather than sourcing them.

The Badge

Modeled in Blender, exported to STL, printed in resin.

↓packat_klart.blend ↓packat_och_klart.stl

Print

Printed on the Anycubic Photon Mono 2 in Craftsman resin for the extra detail. In hindsight standard resin would have been fine at this size.

Scout Buckle Clone — Parametric for Casting

Tue, 02 Jun 2026 00:00:00 +0000

Clone of a Scout buckle — built in Blender Python for casting rather than FDM printing. The goal is a dimensionally accurate reproduction suitable for resin or lost-wax casting.

What makes this different from the rack brace: the buckle geometry is not a simple boolean grid. It has compound curves, a tongue mechanism, and tolerances that matter for function. One-shotting it in Python is harder — the script has gone through multiple partial rebuilds rather than a single config-block iteration.

Scripts

↓buckle_v1.py ↓buckle_part1.py ↓buckle_part2_v1.py ↓buckle_part2_v2.py

The split into part1 / part2 reflects the geometry breakdown — building the frame and the tongue separately before joining, which turned out to be easier to iterate on than a single monolithic script.

What the iteration looks like

Notes to follow. The short version: parametric geometry that has functional constraints (a tongue that must seat and release under load) fights back more than decorative geometry. Changing one dimension propagates into several others in ways that aren’t obvious from a config block alone.

Casting considerations

Notes to follow. Draft angles, wall thickness minimums, and sprue placement are constraints that don’t exist in FDM — the script needs to know about them.

Status

Work in progress. Geometry is close; casting prep not started.

See also: Rack Support Brace — simpler parametric baseline using the same approach.

3D Printed Rack Parts

Sun, 12 Apr 2026 00:00:00 +0000

Filling gaps in the rack build with printed parts — ears, blanks, and a modular tray system. Mix of sourced models from Printables and geometry scripted in Blender Python.

Sourced Models

Model	What it is	Status
1U Universal Rack Ears	Rack ears for gear that ships without them	Printed
1U–4U Spacer / Blank Panel	Blank panels to fill empty rack units	Printed
RackMod 1U Slide-A	Modular 1U tray system with slide-in modules	Testing
19" 1U/2U/3U Cover Plates	Solid cover/blanking plates for 19" rack	Queued
1U Rack Cable Ears	Rack ears with integrated cable management	Queued
Raspberry Pi 1U rack mount	1U mount for RPi in 19" rack (BIFROST)	Printed
Zip Tie Clip 45mm T-Slot	Cable management clips for 45mm extrusion	Queued

Scripted Parts

Geometry generated by a Python script running inside Blender rather than modeled by hand. The immediate need was a support brace for the rack — a 220×150×8mm plate with 40 through-holes in a specific alternating-spacing pattern, the B.E.S.T label engraved into the top face, exported directly to STL.

Writing it in Python means the layout lives in a config block at the top of the file. Changing hole count, plate dimensions, or row spacing is a constant, not a modeling operation.

Rack Support Brace — 220×150mm plate, 10×4 hole grid, engraved text, automated renders | script

More on the approach: Blender Python for 3D printing

Print setup: 3D Printing — Garage

3D Printing

Sun, 12 Apr 2026 00:00:00 +0000

Two printers: an FDM machine for structural and functional parts, a resin printer for detail work. Different tools for different jobs — the resin produces sharper geometry at the cost of more process overhead.

FDM — Anycubic Kobra X

Current machine. Workhorse for rack accessories, enclosures, and anything that needs to be durable and dimensionally accurate. Printing in PLA for most jobs.

Replaced an older Prusa i3 MK0 that still works but is no longer the daily driver. Shelved for now. CNC conversion or rebuild is somewhere on the list, parts donor if it comes to that first.


Build volume	260 × 260 × 260 mm
Bed	PEI spring steel, max 100°C
Nozzle (stock)	0.4 mm hardened steel, max 300°C
Speed	300 mm/s recommended, 600 mm/s max
Extrusion	Direct drive
Leveling	LeviQ3.0 auto-leveling
Multicolor	4-colour native (ACE 2 Pro), expandable to 19
Extras	AI spaghetti detection, HD camera, filament runout

Nozzles on hand: 0.4 mm (stock), 0.25 mm (not tried yet). Expandable to 0.6 / 0.8 mm.

Filament on hand

Brand	Material	Colour	Qty	Notes
Verbatim	PLA	Black	1 kg	Original stock
Bambu Lab	PLA Basic	Orange	1 kg
Bambu Lab	PLA Basic	Green	1 kg
Bambu Lab	PLA Basic	Magenta	1 kg
Bambu Lab	PLA Basic	Clear	1 kg
Bambu Lab	PLA Basic	Red	1 kg
Bambu Lab	PETG	Red	1 kg	Refill spool
Bambu Lab	PETG	Blue	1 kg	Refill spool
Bambu Lab	PETG	Black	1 kg	Refill spool
SUNLU	PLA	Black	0.25 kg	Sampler pack
SUNLU	PLA	White	0.25 kg	Sampler pack
SUNLU	PLA	Grey	0.25 kg	Sampler pack
SUNLU	PLA	Red	0.25 kg	Sampler pack
SUNLU	PLA	Light Blue	0.25 kg	Sampler pack
SUNLU	PLA	Light Yellow	0.25 kg	Sampler pack
SUNLU	PLA	Green	0.25 kg	Sampler pack
SUNLU	PLA	Orange	0.25 kg	Sampler pack
SUNLU	PLA	Pink	0.25 kg	Sampler pack
SUNLU	PLA	Lavender Purple	0.25 kg	Sampler pack
SUNLU	PLA	Brown	0.25 kg	Sampler pack
SUNLU	PLA	Olive Green	0.25 kg	Sampler pack
SUNLU	PLA	Oak	0.25 kg	Sampler pack
SUNLU	PLA	Skin	0.25 kg	Sampler pack
SUNLU	PLA	Transparent	0.25 kg	Sampler pack
TECBEARS	PLA Matte	Black	10 kg	High-speed rated (600 mm/s), bulk stock

Resin — Anycubic Photon Mono 2

Mono LCD resin printer. Used for detail parts — scout badges, finer geometry — where FDM resolution isn’t enough. Paired with an Anycubic Wash & Cure 3 for post-processing.


Build volume	143 × 89 × 165 mm
Screen	6.6" Mono LCD, 4096 × 2560, ~2000 hrs
XY resolution	34 μm
Z accuracy	10 μm (single linear rail)
Print speed	≤ 50 mm/hr
Leveling	4-point manual
Light source	Parallel matrix
Build platform	Laser-engraved aluminium alloy
Data input	USB Type-A 2.0

Wash & Cure 3


Wash capacity	Fits Mono 2 build plate
UV wavelength	405 nm
Cure time	~2–3 min

Resin on hand

Resin	Type	Colour	Qty	Notes
ABS-Like V2	ABS-Like	Black	~3 kg	Structural / strength
ABS-Like 2.0	ABS-Like	Beige	8 kg
ABS-Like 2.0	ABS-Like	Translucent Green	1 kg
ABS-Like 2.0	ABS-Like	Clear	1 kg	For lens work eventually
Standard V2	Standard	Black	~0.5 kg	Display / detail
Standard	Standard	Light Beige	8 kg
Standard	Standard	Translucent Green	1 kg
Standard	Standard	Clear	1 kg	For lens work eventually
Craftsman	Detail	Grey	~1 kg	Sharp detail, brittle

Detailed resin mixing notes and maintenance log kept separately.

Slicer / Workflow

Notes to follow — slicer setup, print profiles, export workflow.

Rack and homelab prints: 3D Printed Rack Parts

3d-Printing on Backend Engineering Strategy Tools

Optics — Can We Print a Lens?

Step 1 — Print and test

Blender Python — Procedural Mesh for 3D Printing

When to use this vs. alternatives

Core pattern

Running the script

Key API surface

Primitives

Transforms

bmesh (direct vertex / edge / face editing)

Boolean modifiers

Joining objects (single boolean cut for a grid of holes)

STL export (Blender 4.x / 5.x)

Collections (organising multi-part output)

Parametric families

Clearing the scene before a run

Viewport layout before export

Working with Claude

Limitations

Related

Procedural Mesh — Blender Python & AI-Assisted Geometry

Stages

Notes

Rack Support Brace — Programmatic 3D Print

Downloads

The script

Step 1 — Render from script ✓

The Printed Part

Resin Badges — Packat & Klart

The Badge

Print

Scout Buckle Clone — Parametric for Casting

Scripts

What the iteration looks like

Casting considerations

Status

3D Printed Rack Parts

Sourced Models

Scripted Parts

3D Printing

FDM — Anycubic Kobra X

Resin — Anycubic Photon Mono 2

Slicer / Workflow