Parsa Ahmadnezhad

building stuff @uwaterloo | alum @tksworldhq

[Parsa Ahmadnezhad @parsaxa]

🛟 Introducing Floaty for @float_financial

> me and @4derekwang went to the Susa x KP event
> thought about panel talk on: Execution > Idea
> decided to execute in 24 hrs
> researched and built an MCP over Float API with 60+ tools

shoutout @anirudhbv_ce

@gtskeg @shaheersan

[Parsa Ahmadnezhad @parsaxa]

@borrasque_

[Parsa Ahmadnezhad @parsaxa]

@UdulaAbeykoon @anirudhbv_ce @nengjiali @Matthew_Wu_ @4derekwang bv thinking about the dams 🦫

[Parsa Ahmadnezhad @parsaxa]

peak life moments:
😀 ‘session limit reset’
🤩 ‘session limit reset’ with this view:

@anirudhbv_ce @nengjiali @Matthew_Wu_ @4derekwang

[Parsa Ahmadnezhad @parsaxa]

@ajay4ai @anirudhbv_ce @nengjiali @Matthew_Wu_ @4derekwang onggg

[Parsa Ahmadnezhad @parsaxa]

@virkvarjun attention is all you need 🔥

[Arjun Virk @virkvarjun]

I just spent months handwriting a 200 page guide on the entirety of ML foundations and math from scratch.

The guide features:
- Neural Nets (Backprop, Adam, SGD, Batch Norm)
- ML Algorithms (SVM, Grad Boosting, K-means, PCA)
- Hardware (Tensor Cores, Systolic Arrays, CUDA)
- Transformers (Multi-Head Attn, KV Cache, LoRA)
- Vision (ViT, Convolutions, MAE, IoU, NMS, VLM)
- Agents (OpenClaw, ReAct, Memory, Orchestration)

Everything I wish I had years ago, for free.

[Parsa Ahmadnezhad @parsaxa]

@lucashjin @gen_intuition clean 🥶

[ani @anirudhbv_ce]

pip install spectralquant ✂️

Up to 6.62x KV cache compression for LLMs and transformers.

Same model. Faster outputs. Smaller KV cache.

Try now (2 mins): https://t.co/7qHRy0DDug

- KV cache integration via @huggingface's DynamicCache
- Three presets: 5.95x (paper), 6.55x (validated), 6.68x (edge)
- Mistral 7B / Qwen 2.5 7B / Llama 3.1 8B verified
- Pure PyTorch + future CUDA kernel support
- Auto-calibration from a bundled corpus

📰 Paper: https://t.co/I89vK1nEFZ

💻 Code, quickstart, and benchmarks: https://t.co/cKJt33vsgk

#LLM #Inference #PyTorch #OpenSource #MachineLearning #LLM #KVCache #Inference

[Parsa Ahmadnezhad @parsaxa]

@daneilp_ @UWaterloo @slashapp dp 🔥

[Daniel Pu @daneilp_]

hey everyone! i'm daniel

- 1st year cs @UWaterloo
- currently swe intern at @slashapp
- 20x hackathon winner
- back-to-back finalist at canada's largest hackathon
- 100+ github stars with an AI sketch to animation tool
- top 99.5 percentile in multiple national math contests
- researched brain MRI segmentation at 16
- #1 kpop fan

new to twitter, but will be posting more about tech, life, and larping. just landed in sf recently, hmu if you want to connect!

[ani @anirudhbv_ce]

This week, we presented STAE-Spectral-Magma at the @RBCBorealis in Toronto.

Our novel hybrid architecture tackles city-scale traffic forecasting by changing the space rather than just the model.

By introducing three parallel spectral views and a bi-axis Mamba state-space model, we successfully surpassed the state-of-the-art @UTSResearch's STAEformer baseline on competitive benchmarks.

We are incredibly excited about the support from the brilliant RBC team. Moving forward, we will be pushing this work as a formal publication and continuing this deep research as an ongoing collaborative project with the organization.

Big thanks to our team: @nengjiali, @UdulaAbeykoon, @enhe_bai and the most excited one here... Ryan LOL! Thank you to our mentors as well who were extremely patient with us during our team calls 😅

#MachineLearning #TrafficForecasting #AIResearch #Mamba #GraphNeuralNetworks #RBCBorealis #TorontoTech

[Parsa Ahmadnezhad @parsaxa]

@anirudhbv_ce @RBCBorealis @RBC R(ocking)
B(v)
C(e)

[julian @phineaose]

WaterlooWorks has a bad ui, no resume matching, no pay visibility,

I built goosehunt to fix that — scrapes Employer Direct via WW's own JS API, scores every posting against your resume with sentence embeddings, parses comp, local FastAPI/Alpine.js UI

[Parsa Ahmadnezhad @parsaxa]

@phineaose master at work

[Prabhdeep @PrabhdeepS_]

Every ANN method either gets slower as your corpus grows or starts retrieving the wrong things.

We built one that does neither.

50K vectors → 10 µs / query
1M vectors → 10 µs / query
5M vectors → 10 µs / query

Introducing MeshRAG: retrieve anything with the same speed, same accuracy, no matter how much data you have.

Here is how we did it 👇

[ani @anirudhbv_ce]

@sentra_app just killed @GoogleResearch's TurboQuant.

Introducing SpectralQuant — 5.95× KV cache compression on Mistral 7B at +7.5% perplexity overhead.

TurboQuant at the same compression: +22%.

3× less degradation. 15-second calibration. One per-model, then drop-in for any HuggingFace LLM, ViT, ESM, AlphaFold Evoformer, or VideoMAE.

📰 Paper & results: https://t.co/I89vK1nEFZ

Check out the findings and how the mechanism works below. ↓

[Udula Abeykoon @UdulaAbeykoon]

Always Saving KV Cache

[Sam Altman @sama]

today we launched ChatGPT. try talking with it here:

https://t.co/uWra8LKFMN

[Parsa Ahmadnezhad @parsaxa]

@austinc808 tuff

[Austin Chen @austinc808]

We killed tutorials.

You no longer have to explain how to do things.

CTRL allows you to record any task and send it as an executable link.

You can finally gain back the endless hours your team was wasting.