I’ve been pumping ideas into a document as a form of productive procrastination, instead of making videos about each idea. I suddenly realized I cannot live without my DGX Sparks, they have become an essential part of my life and there’s also no alternative in the same price range.
So I wondered - what are other people doing with their Sparks?
At work I’ll ask people “So What?” - it’s an obnoxious and jarring question. “Running an LLM”, “Training”, “ComfyUI”, “Fine-Tuning”, “CNN Training”, etc.. those are technical, but, so what? What problems are you solving with your Sparks?
Personal Admin & Documents
Personal document instant recall - e.g Passport/ID expiry tracking. I have everything from birth certificates, marriage, passports, driving licenses, essentially everything about my family. I upload a scan, data extracted, instant recall of any piece of info when I need it - makes things like online check-in for the whole family, a breeze.
Receipt tracking and tax prep (including OneDrive pipeline) - fully captured, extracted, entered into a CSV, private
Family medical records vault
Insurance policy quick-reference
Travel document kit
Bank statement parsing
Voice-First Productivity
Voice memos → actionable notes - instead of voice memos accumulating on my phone, they are transcribed, added to my KB and are forever available.
Multilingual voice translation - obviously available on many devices - but this is private.
Coding & Development
Building apps from natural language - I built a passport photo booth app as part of a visa process, for my entire family. 2 passport photos now cost < $0.50 to create and print. Field tested!
Quick prototyping - basically knock out any idea quickly to explore it.
Code review & debugging / Script automation - the “So what” of this is just far too extensive - it’s probably the heaviest load on the Sparks.
CRM & Contacts
Chat-based CRM from LinkedIn + phone contacts - instantly take notes and schedule follow-ups
Document Intelligence
Contract clause extraction - store contracts and query them in plain language
Resume/CV gap analysis - including passing in JDs to prep custom resumes and cover letters.
Technical paper summaries - cross references and again stored for retrieval and querying
Legal document simplification
Memory & Knowledge
Personal knowledge base - instant recall of facts, decisions and other important items that I’d otherwise have to hunt through folders for.
Research synthesis - from my own written works, I can research and store results for further development
Creative
Game asset creation (Comfy) keep generating with random seeds, essentially for free.
Portrait generation (Comfy) saved me $$ on having proper headshots done.
Voice cloning - my openclaw speaks back to me in a soothing Irish voice.
Home & IoT
Smart home commands
Recipe/meal planning, including low-carb food identification
I like this! Lot of good ideas. How do you keep this all secure? Does your data live on your sparks? I have found that all this stuff eventually snowballs to the point that it simpler to just stay unorganized, lol. LLMs actually now encourage it even more. I have them run my extensive homelab, instead of having a dashboard and ansible tower/semaphore setup I just tell agent I want some app or service to be deployed according to our devops repo and skills and off it goes, underneath it’s still ansible and terraform but the user interface is basically text and gitea issues
Some of the items run directly in folders on my laptop that are MS OneDrive folders, so they’re naturally backed up. e.g. my business accounting, research etc. (I use pi coder for more than just coding)
For all the most personal stuff, data is on the sparks, committed to git and pushed to a raspberry pi with a 1TB NVME drive - I rely heavily on tailscale (which is even on my phone) - so nothing private leaves my network.
I agree on the ‘disorganization is encouraged’ point - I do actually spend quite a bit of time tidying up as information tends to go all over the place.
These are all great ideas and thanks for your sharing!
I used it mostly to handle sensitive documents that cannot go cloud, like university exam papers. I also used it for RAG too to answer queries from colleagues based on SOP and university documents.
Another use is to transcribe long audio, and that is difficult to do with high quality and with long audio even with API, so I prefer ro host my own ASR service. In my language it is hard to transcribe accurately enough, and so I have to use two to three ASR models to transcribe, and ask LLM to merge them for me.
I also use it to produce podcast for my own learning.
In my personal life, it is still more like an experimental tool to test out new models, especially those not available on API. I feel like Sparks give me the freedom to explore new models and new settings etc. With API, I often feel that their quality changes from time to time (perhaps they adjust their setting and quantization), and so Spark gives me more confidence about quality consistency.
A few decades of e-commerce SAAS has kind of gate-kept the most important business tools like CRM and basic book keeping has relatively been solved with 1 spark and the right harnessing. In general, the long context ability and overall needle-retrieval precision for MBA toolkits are daunting to say the least. Spark has the perfect architecture and specs to actually handle that, while being the size of a stack of CDs and takes less power per month than a playstation.
Home assistant replacement using qwen as the master controlller ai, music dj on the go, stock trading ai during market hours, engineering teacher ai for the kid, personal assistant for me and wife, budget updater, frigate hook up to ai to monitor house, comfyui, etc… I can do this forever. Sadly mine went to RMA due to power sensor problems and has not come back yet… :( I just got laid off and I’m losing my mind wanting to work on ai again.
I actually find that organization is something agents can do for you. I have a list of organization criteria that Claude Code can bring up in a skill, and when things get disorganized - I just ask Claude code to tidy up, and things go where I expect them to be. I have a list of folders it’s not supposed to look in, a rule that it uses my inference stack to determine categorization of contents of ambiguous files instead of looking at them itself (so it doesn’t read secrets or private info) and instructions on my folderization practices - when why and how(to go with the existing wheres).
So, private file investigation as a tool for my agents that are served by the cloud.
I also use my spark for email triage, for the same privacy purposes.
I use it for development when i need to burn oodles of tokens that don’t have to be smart, like I have a card game where it plays both sides so I can find bugs and develop strategy guides for the decks(for the llm).
I built a narrated dungeon explorer, super basic, think a text rogue, but if they quit working on it after a day, then wrote a zork style interface for playing it. So the llm powered narrator describes everything that happens, which is pretty cool, but what’s even more awesome is that it randomizes themes and backgrounds for the setting and characters, so you get an entirely different kind of story every time. One of my earliest AI projects, that I should probably put more time into. it’s kinda awesome. Just.. super niche? Like who wants to play a zorklike in this day and age?
Erotic roleplay. Don’t look at me like that.
I have a project where I’m trying to create a harness for emulating a person over a lifetime. Experiential memory, recognition, relationships, knowledge collection, derivation, etc. Very much a wip.
TTS. who needs audiobooks really? and stt, because sometimes typing is just too hard.
Super high resolution maxed out settings Steams Games. Check out FEX-emu… pretty neat.