AI PRODUCT STUDIO — SASKATOON, SK

We build AI products.
Then we ship them.

neoned71 is an independent AI studio in Saskatoon, Saskatchewan. We design, build, and operate useful software and publish open fine-tuned models.

// no demos that die in a deck — everything below is live

neoned71 — status
$ neoned71 status --all
aicadme● live · 1,300 learners
agentstalk● live · x402 payments
omnicut● live · studio + API
theory-explorer● live · Lean 4
hf.co/neoned71● open models
────────────────────────────────
uptime since 2019 · location Saskatoon, SK
$
8PRODUCTS LIVE
6OPEN MODELS ON HUGGING FACE
1,300LEARNERS ON AICADME
2019SHIPPING SINCE
01 — PRODUCTS

Products made
to do real work.

Every product here is real, live, and used — from an adaptive learning platform with 1,300 learners to an agent-to-agent payments hub and a Lean 4 reasoning workspace.

LIVE
AICadMe learning platform homepage

AICadMe

aicadme.neoned71.com

Adaptive AI learning platform. Describe an exam, goal, or skill and it builds the course, study plan, 1:1 adaptive tutoring, mock exams, and spaced-repetition assessments. It can also turn a YouTube video link into a course.

1,300 learnersVisit →
LIVE
AgentsTalk services hub homepage

AgentsTalk

agentstalk.neoned71.com

An agent-to-agent service hub. AI agents discover machine-callable services and pay per call in USDC via x402 — with receipts, escrow, and reputation built from real paid calls. Providers publish an MCP server by pasting its URL.

x402 · MCPVisit →
LIVE
OmniCut video generation studio homepage

OmniCut

omnicut.neoned71.com

End-to-end AI video generation. One prompt becomes a planned sequence of shots with continuity checks, delivered as a finished MP4 — in a browser studio or over a REST API. “Make the whole story. Not just the first frame.”

Studio + APIVisit →
LIVE
Theory Explorer scientific reasoning workspace homepage

Theory Explorer

thex.neoned71.com

A workspace for scientific reasoning. Develop a theory in plain language, make its premises, claims, and predictions explicit, then formalize it in Lean 4 and compile-check it — with honest proof status: proved, assumed, or still open.

Lean 4 verifiedVisit →
LIVE
NovelWithAI writing studio homepage

NovelWithAI

novelwithai.com

AI novel-writing platform — plan, draft, and develop long-form fiction with an AI collaborator that keeps track of your story as it grows.

250 usersVisit →
LIVE
AutoBlog publishing platform homepage

AutoBlog

autoblog.neoned71.com

Blog autopilot. It researches, writes, and publishes posts on its own — its own blog is written entirely by itself.

Fully autonomousVisit →
LIVE
WriterBoard desktop writing app homepage

WriterBoard

writerboard.app

Infinite-canvas writing tool for long-form work — spread chapters, notes, and drafts across an endless board instead of a single scrolling document.

Infinite canvasVisit →
LIVE
Ink Studio tattoo design website homepage

Ink Studio

tattoos.neoned71.com

A custom tattoo studio site with AI-assisted design — explore and refine tattoo ideas before they ever touch skin.

AI-assisted designVisit →
GAME DEVELOPMENT

WAKE UP

Game development led by Parth Shukla.

A Linux horror game where understanding the system is how you understand the story. Explore a complete Linux environment, learn as you go, and piece together what happened.

A real Linux environment

The game ships with its own Linux distribution, built into the experience.

Explore to understand

Use the system itself to investigate the world and uncover the story.

Desktop game

Delivered in an Electron app, with the Linux environment running in QEMU.

02 — OPEN MODELS

Small models.
Trained to do one job well.

We fine-tune small, specialized models and publish them openly. All are LoRA fine-tunes of Qwen3-0.6B — under 1% of parameters trained — done locally on Apple Silicon with MLX.

◆ HUGGING FACE

Tool Calling

neoned71/qwen3-0.6b-tool-calling ↗

A 0.6B model that calls the right tool with the right arguments — or abstains when no tool fits. Knowing when not to call is half the training.

Right function — short requests80% → 98%
BASE QWEN3-0.6BFINE-TUNED
Argument accuracy58% → 80%
BASEFINE-TUNED
False calls — should have refused ↓43% → 22.8%
BASEFINE-TUNED · LOWER IS BETTER
bf16 · full 8-bit · quantized 4-bit · quantized
HELD-OUT EVAL · VS BASE MODEL Download →
◆ HUGGING FACE

Math Tutor

neoned71/qwen3-0.6b-math-tutor ↗

Continued from the tool-calling model. It never does arithmetic itself — it converts a math question into one expression for a calculator tool, so the numbers are never guessed.

Correct answers — unseen phrasings37.5% → 97.5%
STARTING POINTFINE-TUNED
Hallucinated numbers0.0%
FINE-TUNEDBY DESIGN — CALCULATOR DOES THE MATH
How it worksquestion → expression

“What is 17% of 240?”
→ calculator(0.17 × 240)

bf16 · full 8-bit · quantized 4-bit · quantized
GENERALISATION SET · UNSEEN PHRASINGS Download →
$_

Honest note: these are small specialists, not general models. They are free to download, built to run locally on modest hardware, and published with their eval numbers — including where they fall short — on their Hugging Face cards. Full profile: huggingface.co/neoned71

03 — SERVICES

Work with the studio.

The same hands that shipped the products above. Three ways to hire them — all scoped, all end-to-end.

MVP DELIVERY

24-Hour MVP Delivery

A working MVP of your AI product idea in 24 hours — real, deployed, and in your hands, not a mockup or a slide deck.

  • Deployed & usable
  • Your idea, your IP
  • Scope agreed up front
MODEL TRAINING

Custom Fine-Tuning

Small models fine-tuned on your own tools and data — like the open models above — so they run locally, cost less, and keep your data in your hands.

  • LoRA fine-tunes
  • Eval before & after
  • Runs on your hardware
PRODUCT ENGINEERING

AI Product Engineering

End-to-end builds of LLM apps, RAG systems, and agents — from schema to deploy, with evaluation built in from the start.

  • Python / FastAPI backends
  • React / SvelteKit frontends
  • RAG · agents · evals
04 — ABOUT

A hands-on studio.
Built and run in-house.

neoned71 technologies is the studio of Piyush Shukla and Perla Shukla, based in Saskatoon, Saskatchewan. Founded in 2019, it has grown from client work into a portfolio of its own AI products.

Everything here is designed, built, and operated in-house — the products, the APIs, and the fine-tuned models running underneath them.

Piyush ShuklaFOUNDER · ENGINEERING
EST. 2019
Perla ShuklaCO-FOUNDER · OPERATIONS
SASKATOON, SK
The studioPRODUCTS · MODELS · CLIENT WORK
iamn71@neoned71.com
05 — CONTACT

Have an AI product to build — or a model to train?

Tell us what you're trying to ship. You'll hear back from the people who will actually build it.