Hacker News Reader

aadyachinubhaion [dead]

Just for fun, I’m experimenting with converting NumPy code to SymPy and applying concolic testing principles.

aadyachinubhaion Claude Opus 5.5

Huh ..., not another one!

aadyachinubhaion Software Sandboxing: The Basics (2025)

For LLM generated Python code, pydantic's monty is awesome: https://github.com/pydantic/monty

Plus, pydantic is funded so it will be well maintained.

aadyachinubhaion I can't stop thinking about Papua New Guinea

The title of this post is super intriguing

aadyachinubhaion Show HN: Pizza Bot – An inbox for AI agents that work in the background

cool idea, how helpful is this in practice?

aadyachinubhaion Trying to Make a Loop Auto-Vectorize

maybe unrelated by NumPy and Cython have made it incredibly easy to write vectorized code.

aadyachinubhaion Ask HN: What are you working on? (September 2026)

https://github.com/aadya940/scikit-verify

Trace Python+NumPy programs into SymPy formulas and check them symbolically, across all branches

aadyachinubhaion Pandas Should Go Extinct

The reality is most of the pandas audience don't care about performance. Whenever performance is in the question people have always used polars, duckdb, dask etc. These users are usually software engineers and not data analysts. Agreed, that there is a real gap in latency and performance though!

aadyachinubhaion Claude is only available to people over 18 years

great, save the next generation from cognitive decline

aadyachinubhaion AI 2027 (2025)

It is speculated that there was context leak from a NYU Professor's chat I think? I contribute to SciPy heavily and I've seen AI shit the bed a few times now. So 13 million lines of lean? Hell no dude!

aadyachinubhaion AI 2027 (2025)

One of the first things I learned in ML is that the model can only learn from the information in X. If X doesn't contain enough information to determine y, no amount of compute can fully recover it.

That's why I'm skeptical of grand claims about AI. Scaling can make models much better, but it can't create information that isn't there. An AI system can be extremely useful without becoming superhuman.

aadyachinubhaion AI 2027 (2025)

I don't like grandiose claims about AI.

aadyachinubhaion Python sets and dictionaries can have quadratic-time performance

That is why I like to use __slots__ when defining a class. Unlike dicts, using __slots__ is a tuple so using it to store class attributes is much faster.

aadyachinubhaion I resigned from Anthropic today

Am I the only one who still believes that these resignations are not a big deal?

aadyachinubhaion Show HN: What if the speed of light was 5 km/h?

Awesome, at 5 km/h it feels like I'm in a black hole. Time stops at the speed up light if I'm not wrong!

aadyachinubhaion Tracing np.add, all the way down

The `.c.src` files with the @arch@ like dispatching logic is specific to numpy. Also, they are currently making efforst to shift to C++ templates using Google's Highway project. I worked on porting `np.negative` to it!

aadyachinubhaion The Navier–Stokes Millennium Prize Problem

LLMs can't contribute good code to some of the good OSS math libraries, How is it even solving these problems?

aadyachinubhaion Show HN: n8n like workflows for AI agents that control a real VM

Each node in the graph is a plain English instruction. An AI agent executes them in order inside a Docker container with a full browser and desktop. Because each node is independent, the agent stays on task, it doesn't drift or hallucinate its way through a free-form prompt.

No programming required to build workflows. If you can describe a step, you can add a node.

I personally use it for:

- Applying to jobs automatically using my saved credentials, exactly the way I would do it manually

- Scraping websites and running data analysis on the results in the same workflow

- Checking my university LMS on a schedule for pending assignments

A few technical details:

- Nodes: Navigate, Do, Read, Fill, Check, Code, ForEach, Bootstrap

- Any LLM via LiteLLM: Gemini, GPT-4o, Claude, Ollama, OpenRouter

- Watch it work over noVNC, pause and take control, hand back anytime

- Chrome sessions persist across restarts via a named Docker volume

- Webhook + cron triggers, secrets vault, human-in-the-loop confirmation per step

GitHub: github.com/aadya940/orbit-ui

Docs: orbit-cua.com

aadyachinubhaion Show HN: Structured Python control over AI computer use agents

Fair, mature frameworks do a lot more. Orbit isn't trying to out-feature them. It's a thinner abstraction specifically for computer-use, where the bottleneck isn't orchestration complexity, it's controllability at the step level. Different problem.

aadyachinubhaion Show HN: Structured Python control over AI computer use agents

Most agents are prompt + a bunch of tool calls. Writing everything in one prompt is like writing your entire app in one function. It works until it doesn't, and when it doesn't, you have no idea where it broke. Orbit gives you steps and python control flow instead of prompts, so failures are local, models are swappable, and budgets are per-action. This makes it debuggable.

For example, Your `Read` step failed after 3 LLM calls on step 4 of 7. With a monolithic prompt, that's just... it hung.

aadyachinubhaion ChainoPy: A Python Library for Discrete Time Markov Chains

Why ChainoPy? Covers most of the fundamental agorithms for Markov Chain Analysis Memory efficient Model saving Faster than other libraries (eg: 5x Faster than PyDTMC) First Package to contain functions to build equivalent Markov Chain Neural Networks from Markov Chains. Contains Markov Switching Models for Univariate Time Series Analysis