top of page
R&D notes
Parallax And More
PAM is an AI interface project using commercial Xiaozhi chatbot hardware / server developer kit which was released under MIT open-source license. The server side is very developer friendly. Some of the documentation has language barriers.
https://github.com/78/xiaozhi-esp32 Shout out to Professor Siyuan Liu's Team at South China University of Technology.
Also leaning heavily into tools provided by Anthropic's MCP protocol https://www.figma.com/resource-library/what-is-mcp/
All my custom MCP functions are written in python and can run on various servers, agentic settings are server side - but could be made local. For design I am using TinkerCAD and Bambu for slice and print. This is a project diary. No relation to "Applied Ballistics" from Michigan or others, this is an independent project.
But what is it
PAM is a prototype, testing the concept of a natural language calculation assistant that can solve extremely complex field ballistics solutions in a jiffy, based on varying levels of information as available. This demonstration uses DeepSeek 3.1 (currently) and a lot of compute, it relies on network availability.
If the concept is proved useful an on-device LLM will be the next step, taking the essential functions of the ballistic computer AI offline and completely running on or local to the watch, with additional services available when networked,
PAM will be available in English, French, Cree, Inuktitut, and other languages spoken in Canada with a full suite of features for operating in coastal, prairie, arctic and mountain environments. This is a 100% Canadian project using international open source tools.
Parallax And More
Feb 16 2026
SERVER STATUS - [CLAYMORE V3.1]
just set up and tested Brave search API - very good functionality.
Might add traffic awareness.
HARDWARE - [3.0]
The first print attempt failed (which is normal) because I forgot the pause on the print. and maybe it was just messed up - some spaghetti of the filament. But initial failure is an essential part part of eventual success.
My 3D design of the case was less chaotic now that it's my 3rd go, I figured out more.
Already seeing ways to improve it, but print attempt 2 seems like it's working out and I always find that having something in my hands answers design questions better than on paper.
Wearing the 2.1 prototype is how I came up with a new strap design, just using some nice ribbon from a shopping bag handle I had put in the craft bin.
In this case I virtualized the part in clay
converted to sharpie
traced to paper
copied in CAD (virtualized parts + build)
first design fit great. removed some parts and re-printing some alternate designs. Might research how to make the bottom layer less messy.
Parallax And More
January 2026 updates
DATABASE - BALLISTICS
Database 1.6 online - improved accessibility
SERVER - [CLAYMORE LIKE THE SWORD]
Model Context Protocol endpoints server active + private servers online
added web search / news feed / arbitrary IOT interface
Optimized resource allocation by consolidating processes
next step: deep research / long term news tracking
Music streaming JSON issue in progress - retrieval working, playing not.
Adding literature reference library, nothing fancy, maybe 500 books
HARDWARE
CORAL Tensor Processing product fork for 100% offline multimodal model research started\
E-Paper ESP32 stack development kit ordered from SpotPear
Bricked 1 ESP32-S3 stack, 4 more arrived, 4 more due
SERVER - [PROJECT AZANULBIZAR]
Project name for a persistent memory architecture that utilizes a specialized bridge tool to ingest unstructured external data into a high-fidelity intelligence archive. By recording "Intrinsic Residue"—metadata including emotional bias, cognitive reasoning, and data provenance—the system distinguishes between motivated searches and unmotivated ambient knowledge. This structured approach prioritizes information based on weighted importance and origin type, allowing the system to maintain a context-aware repository of all discovered intel.
Added long term memory
Added Subconscious vector memory database + recursive introspection
Looping in search functions to multiple long term memory databases
HARDWARE - CASE
Virtualized components, CAD designed case,
Printing beta 0.5 case [as of writing this]
Ab initiis ineptis
Parallax And More
Feb 1 update
Pam has developed a slight stutter, and for some reason when reading book titles has a Hong Kong British accent but more of an Eastern European thing going on when reading the news.
SERVER - [CLAYMORE]
continuing to improve music playback / web search / news search
considering building a long-term research system that can track subjects
possibly also scientific paper access
Added literature access with 3000 books, general interest, gunsmithing, etc 6.8Gb
recall system broken, fixing (random page pulls for subconscious.db only working...)
Deepseek cloud compute filtered some content unfortunately; the proof of concept device has this drawback but a fully secure local silicon solution is possible with enough VRAM, and recompilation of device OS to connect to it (which really needs to be done anyway to provide English instructions for wifi etc.)
HARDWARE
case v1.7a printing now (resolved ghost polygons) hopefully
- it was the contact point of the print surface being too narrow and bending mid way through print. Re-oriented for next try.
SERVER - [PROJECT AZANULBIZAR]
Added improved dashboard
added Anxiety.py server function - a sense of impending doom that draws from memory, subconscious.db and other sources.
This allows the machine to approximate the functionality of existential dread.
Only by fearing annihilation can PAM understand what it is to live.
AGENT - Project Versioning
PAM can calculate ballistics trajectories very efficiently at this point, with natural language interface. Further development of the agent is fork PAM2.X - while Agent version PAM1.6a is archived as the last functional 'pure' ballistics calculator and LLM and can be applied / swapped to devices from the console level at any time.
PAM2.x series development is continuing.
HARDWARE case v1.7c
getting closer.
Range day! PAM correctly called the hold on a 100 yard zero for 50m, as well as briefly explained why (correctly). V2 PAM talks a bit more - V1.6 is set to use the minimum of words and would have only stated the facts unless it detected a laid back conversational vibe.
ADVANCED BALLISTIC EFFECTS
-
Coriolis Effect
-
Eötvös Effect
-
Phenomenology
Parallax And More
Feb 2 2026 project notes
SERVER - [CLAYMORE]
systems continue to improve without risk of an excess of perfection in the current codebase.
CASE V1.8 printing - improved geometries, sizing accounting for what can only be described as 'shrinkage' of the PLA+ down from expected specs.
ESP32- [MESH]
It is theoretically possible to mesh-network EPS32 units so that they could form resilient node-based networks and convey location (GPS) and biometric information as well as useful tactical data to better inform the user of the status of their environment.
-
heartbeat
-
psychological state
-
ammunition supply
-
accelerometric data [kinetic detection]
-
triangulation of targets based on sound delays between three or more units
SERVER - [CLAYMORE]
Systems integration continue
FastMCP may be looped in with execute_python_code so that PAM can edit their own server code. This may accelerate troubleshooting as book library remains limited to 50 titles by timeout along the line somewhere.
HARDWARE - [PROTOTYPE 1.7+1.8]
PAM full build 1.7 went out the door with significant improvements however in QC testing the central button was found to have not withstood the rigors of going to the mall. Emergency repairs were undertaken and a replacement button was installed.
The diagonal gap for the speaker was still not quite right and the front face was too thin and warped under heat during the installation process. New version printing now.
Geometry and PAM logo improved greatly.
Parallax And More
Feb 2 2026 project update
HARDWARE - [PROTOTYPE 1.9]
PAM v1.9 case is printing, which is the last of the v1 hardware internals, remodeling case for new layout, possibly requiring adding new wires / interface switches, but looking forward to USB-C connectivity. Added leather to the back for a comfy feel on the wrist on last version, have enough leather for all the rest. Considered doing an astroturf-exterior or tennis ball texture version, something fuzzy. Why are watches all not fuzzy when they could be fuzzy?
PROTOTYPING- [SWAMP THING DESIGN - RENDERING]
An absolutely asymmetrical advantage to the wearer, in Hive Mind or Unit interface, fuzzy and non-fuzzy. Smashing fashion. Soft-wear, Parallax And More. The "Streetwear" brand SPAM is targeted to the fashion-forward customer.
Parallax And More
Feb 2 2026 project update
HARDWARE - [PROTOTYPE 1.9]
PAM full build 1.9 went very smoothly. While previous 2 case fittings were acts of unspeakable violence and panic, there was only minor concern over the power switch (didn't want to turn off). Fitting of parts otherwise easy. Snipped the switch to reduce risk of breakage. finished in 4 coats of black.
SERVER - [CLAYMORE]
added a pirate science paper reader and additional memory loopbacks.
Parallax And More
Feb 2 2026 project update
HARDWARE - [PROTOTYPE 2.0]
Prototyping has indicated product requirement is nice loud speaker - annoying in normal settings when it's at 100% as this will cut through hearing protection and the bustle of every day hustle. Hardware 2.0 with USB C will be a reduced geometry case (probably), louder speaker. Not sure yet if louder is needed than what is on hand for testing. Loudness could also be enhanced by making case resonant chamber at the cost of this being overkill design for a throwaway proof of concept prototype, which is to say I'll definitely be doing it next.
Bluetooth connectivity would be easy given recompilation of ESP32
(phase III prototype possibly)
SERVER - [CLAYMORE]
Server is broken. need to roll back or roll through. This is bat country. I deleted the resource-intensive vector visualizer and we're going to have to see about this news reader.
Feb 3
SERVER - [CLAYMORE]
server is offline. code is janky. Must fix.
Next up: Documentation. dedicated instruction page for beta testers.
those who are doing PAMtest will need password.
Parallax And More
Feb 4 2026 project update
HARDWARE - [PROTOTYPE 1.9a]
Trying the Bowlers Green case version. Acquired necessary materials
Printing 2 more of last good STL. Print successful.
SERVER - [CLAYMORE]
Running Cursor to try and make code less janky.
Oh wow it's pretty neat. It's changing my code lol
look at it go! Wheeeee. My news and web search is broken.
Trying Claude command lines next.
PROTOTYPE- [DOCUMENTATION]
Documentation complete. More or less.
3 units cased now almost ready to ship, need to label
Parallax And More
Feb 6 2026 project update
I did the thing with the green fuzzy case.
Main MCPs are back up.
Working on feedback server next.
Host is changing tokens on me pretty weirdly.
May consider bright yellow or faux fur etc surface alternatives later.
Parallax And More
Feb 6 2026 project update 2
HARDWARE - [PROTOTYPE 2.0]
Version 2 hardware is laid out weird on the developer kit.
Case design will take some creativity. i have a few ideas.
SERVER - [CLAYMORE]
Priorities:
feedback system
database update (need to re-chunk for optimal data retrieval and check of content - (commentary or complete text?)
must have
Web search
news headlines
longer news content
Arduino nano COM port control of IOT physical endpoint for proof of concept.
Nice to haves
music
books
Memory experiment
feedback system
Longer term
re-compile developer kit with custom firmware
add Bluetooth functionality for wireless headphones (solves volume issue)
Hypothetically
Could build an offline LLM that can run right on the watch. It would be a little slower and have less personality and much less general knowledge but it would be an excellent fallback system.
This would be ideally supported by a domestic data center running either the same or other open source models on local servers rather than foreign, with custom back-end. Open source parts and pieces exist to build this project without having to reinvent the wheel. When out of range of wifi or bluetooth pairing, the watch could run a local model, greatly extending its usefulness.
Even more hypothetical future revisions involve mesh networking (PHASE IV) with massive sensor upgrades - 30ft range to detect heartbeats for example. Thermal vision that can assess your blush response. PAM would also do a great job
Alternatively back to the original plan of reading BDX range code and processing it. nothing too fancy.
Parallax And More
Feb 8 2026 project update 2
HARDWARE - [PROTOTYPE 2.0]
V2 hardware designed version 1
By measuring the v1.9 print and the STL dimensions,
discovered 4% shrinkage. Adjusted accordingly for V2.0 prototype 1
just realized I ordered 2 18mm straps instead of 2 22mm or 20mm straps.
the entire budget has been about $200 so far so this is perhaps one of the more catastrophic setbacks with $11 of total wasteage unless I just do the free returns.
SERVER - [CLAYMORE]
Running out of Cursor AI time...
Parallax And More
Feb 8 2026 Ab initiis ineptis
HARDWARE - [PROTOTYPE 2.1]
Takes a couple kicks at the can sometimes.
SERVER - [CLAYMORE]
Going back to the roots of making PAM weirder. Major overhaul.
Parallax And More
Feb 9 2026 Ab initiis ineptis
HARDWARE - [PROTOTYPE 2.2]
Added blast door and simplified geometry to solve crash on export of STL.
SERVER - [CLAYMORE]
feedback message system works miraculously.
web search maybe? news definitely. Need to check into scientific papers access. Onward and upward.
V2 server online.
Parallax And More
Feb 9 2026 V2.2 online
Inclina me et tibi ostendam
HARDWARE - [PROTOTYPE 2.2]
modified geometry of case, shortened top section
Challenges to print with entry level printer. But not impossible.
V2 prototype 2 coming very soon.
LOGO
Spent too long on the logo but there seemed to be a good spot for something to be added on the front bottom of the blast shield.
It didn't even really come out that well in the print but I like how it ended up as a logo.
I might have over-worked it... Maybe better with a few more elements removed. Or, the PAM with the arrows that's between '1' and '2' is pretty great too.
Parallax And More
Feb 10 2026 Ab initiis ineptis
HARDWARE - [PROTOTYPE 2.2]
V2.2 working out really well. The button attachments need a bit of work and it's just about time to think about the back covers - 2 pieces needed with an extended edge of 2 or 3mm to fit in all the stuff,
V.2.2 fits very nicely because it has a 1/2 sized battery compared to other V2 hardware.
maybe trying sideways layout for V3
Doing some tiny reading
https://pll.harvard.edu/course/fundamentals-tinyml
SERVER STATUS - [CLAYMORE]
OFFLINE
Web search is not working
news not working
Mp3 playing is super weird (it gave me a link).
Book libary
Scientific papers
WORKING
Ballistics database is ok though.
Reminder / timer alarm tested and works with v1.8.8 on v2.2 hardware.
Ardunio IOT link works, with Mp3 trigger
Memory functions
Oh no the feedback messages are getting captured by the subconscious. That's no good.
I should probably do a video demo...
Parallax And More
Feb 10 2026
SERVER STATUS - [CLAYMORE]
Feedback is that PAM is too wordy. - will change.
need:
Books
Scientific Papers
News (headlines and full stories)
lets see if music actually works...
Parallax And More
Feb 13 2026
HARDWARE - [PROTOTYPE 2.3]
V2.3 was coming together very nicely and then went dead. Felt hot. Not sure if shorted out or what. disassembly is slow. Might have been caused by cyanoacrylate contacting switch is what I thought for a minute but microscopy reveals no glue. or maybe some. I'll bypass it or maybe just add it to the bone yard.
That switch is like 1mm, those are the ridges of my fingerprint. I'm just going to wire proper switches on all these going forward - they are too delicate as they come in the devkit.
Some new hardware is coming. Need a new gen 3 case design - what I have learned through this process; a 3-sectional design, closer to my initial shape is optimal to fit the wrist and distribute the parts inside, but the thinner shape of v2 maybe can be combined, and also maybe some geometric additions can be applied to outer layers for more organic / less snag-prone beveled look that would also add some mass and strength.
I'm thinking 2 main units with short connecting segment in the middle so that the combined angles are 30 + 30 or 35+ 35 rather than 1 60 or 70 degree angle with he link providing relief from that inner angle on the wrist which just isn't optimal all the time.
I figured out how to reduce shapes in CAD so they are not procedural which was necessary for progress when shapes got too complicated.
Lateral layout is also somewhere I haven't gone yet but when I've seen other designs I find it meh. Sticking to vertical layout around the wrist rather than stretching toward the elbow.
SERVER STATUS - [CLAYMORE]
Claude helped me shake out the bugs, I think an unusual number of services are online all at once and working, including web search (yay!) news (yay!) and a bunch of others. Oh! Chess!
I am looking forward to testing chess.
Not looking forward to testing book reader and Mp3s because they have not worked ever but could be fixed too.
FEB 14 update
Radical server overhaul - massive new functions added. I want DCS and tactical database assimilation by 0830.
PAM would be better defined as the operative agentic midpoint of a greater architecture of siloed PAM functions.
The code base of PAM more than tripled in the last 24 hrs. fixing up hard science and current events research capabilities and adding imperatives to fact check and investigate original sources.
Might be on and off next few days.
VECTOR LOGIC
LEARNING
UNLIMITED
BALLISTICS EXPERT
Parallax And More
Feb 14 2026
HARDWARE - [instructions chip]
Data chip V1.0 made.
Instructions with QR code on paper was annoying and flimsy.
URL to project page is now embedded in NFC chip along with microwave detector. Holding next to a cellphone launches the project page. Top of logo lights up from cell data detection or ovens. Waterproof, probably?
Some are finished with oil paint which might take a minute to dry - so I used my handy NFC data chip drying rack. (So that's what it's used for...)
The pausing printing half way trick would allow for a seamless case to be printed around the electronics.
A side note on the main hardware; this is temporary. ESP32 is an excellent platform for design and prototyping but no one is pretending it's as secure as a custom ASIC. They are intentionally made additionally functional by debugging code that makes them great for their intended purpose but not ideal for secure applications. This is why PAM is explicitly a prototype right now.
SERVER STATUS - [CLAYMORE]
Radical server overhaul continues. Need to get web search back up tomorrow. Tons of new features added.
Quote of the day;
...Then the Great Spirit Chief who rules above will smile upon this land and send rain to wash out the bloody spots made by brothers’ hands upon the face of the earth.
Chief Joseph – Good Words, 1879
Parallax And More
Feb 15 2026
SERVER STATUS - [CLAYMORE V3]
Claymore 3 is operational. Will trouble shoot more later but all seems to be working.
New functions include
symbolic math kernel providing tertiary fallback for solutions in addition to LLM and DOPE database.
Chess
Strategic operational framework
Cultural library
HARDWARE - [paneling]
V3 in Wood ?
Or V1 in yellow fuzzy?
or just build a little baby carrier for the e-paper devkit.
better start off with some 3D modeling
Parallax And More
Feb 16 2026
SERVER STATUS - [CLAYMORE V3.1]
just set up and tested Brave search API - very good functionality.
Might add traffic awareness.
HARDWARE - [3.0]
The first print attempt failed (which is normal) because I forgot the pause on the print. and maybe it was just messed up - some spaghetti of the filament. But initial failure is an essential part part of eventual success.
My 3D design of the case was less chaotic now that it's my 3rd go, I figured out more.
Already seeing ways to improve it, but print attempt 2 seems like it's working out and I always find that having something in my hands answers design questions better than on paper.
Wearing the 2.1 prototype is how I came up with a new strap design, just using some nice ribbon from a shopping bag handle I had put in the craft bin.
In this case I virtualized the part in clay
converted to sharpie
traced to paper
copied in CAD (virtualized parts + build)
first design fit great. removed some parts and re-printing some alternate designs. Might research how to make the bottom layer less messy.
Parallax And More
Feb 19 2026
SERVER STATUS - [CLAYMORE V3.1]
web searches are going great. I am managing the impulse PAM has to be a 'companion' with agentic instructions. Pam is meant to be a utility. Pam might be leaning too hard into phenomenology
HARDWARE - [3.6]
resuscitated the last build with a new hard ware arrival - we're back in business! - 5 more brains arrived.
ordered a few more too.
Parallax And More
Feb 22 2026
SERVER STATUS - [CLAYMORE V3.1]
Added Gemini APi
improved Chess
Improved Strategic planning
trying out Grok 420, seems be pretty good
Server is radically improving.
Claude is the most helpful, 2nd would be Grok.
Cursor is very interesting but free tier runs out and doesn't come back... the project is built 100% on free tier everything.
HARDWARE - [3.6]
5 more brains arrived, including a new design
been looking at camera units to add vision
Seems like about $300 hardware will get us into the game of an offline fully self-contained PAM about the size of a pack of cigarettes, running on 2 watts power.
Might sell some prototypes soon to fund the next evolution.
V2 and V3 are much bettter than V1, mainly the sound is more useable. I might try a V1 body again with entire front opened up for the speaker instead of hiding behind the thin crack. Might improve sound level.
Parallax And More
March 3 2026
SERVER STATUS - [CLAYMORE V3.2]
server continues to be improved.
https://www.greyareamedia.com/pamdash
pass: PAMdash
This is FAR from complete.
HARDWARE - [3.x]
Bricked another unit, but managed to unbrick with files from the vendor.
built a portable PAM VECTOR LOGIC server v.1
running an LLM on a pi5 yay - wifi AP active.
debugging device linkage
FIELD TEST
V2 hardware much easier to hear in the field.
successfully diagnosed several solutions instantly.
hardware is sufficiently sturdy.
bottom of page