Introducing myself
Hi everyone! 👋
I’m Rajiv, and I’ve been working on a project called tseda (Time Series Explorer).
If you’ve ever stared at a messy time series dataset and wondered where to even start, I built this for you! It’s an app designed to take the headache out of exploring and decomposing regularly sampled data (hourly or greater).
The "Why" behind it:
I wanted a tool that didn't just give me a raw plot, but actually helped me understand the structure of the data. tseda uses Singular Spectral Analysis (SSA) to break things down into Trend, Seasonality, and Noise, but it does the heavy lifting for you—automating window selection and suggesting component groupings so you aren't just guessing.
Some cool bits:
Suitability Checks: It’ll actually tell you if your data is "too noisy" for SSA before you waste time on it.
Change Point Detection: It automatically flags trend shifts and seasonal jumps.
Notebook Parity: You can use the Dash UI for quick exploration or the Python API if you’re living in a Jupyter notebook.
It’s built with Python 3.13, and I’m always looking for feedback or new ideas to make it even more useful for the community.
Check it out here: https://github.com/rajivsam/tseda
Replies
i have absoloutly no clue how to use git, can you kindly tell me how i can see this in action
@bhavyabafma I need to work on video demonstrating that, definitely on my todo list, as soon as its ready, I will update this thread. Thank you
Super neat project! Time series decomposition without the guesswork...love it. I will check out tseda. If you're up for it, I'm launching The Sponge on PH soon...an AI-powered flashcard app that turns any webpage into study material with spaced repetition. Would appreciate a follow (See "PRODUCT HUNT LAUNCH" Link in my profile).
@rianbrob Thank you Rian, cool - maybe I can use that to describe algorithms for tseda and other stuff
Hey Rajiv — this is really interesting.
I’m not deep in time series analysis myself, but I love the idea of a tool that doesn’t just show you the data, but actually helps interpret what’s going on underneath it. The suitability checks and change point detection sound especially useful because that’s where a lot of people probably get stuck — knowing whether the data is even worth analyzing in a certain way.
I’m building in a different space with Traction, focused on helping small businesses turn visibility, content, leads, and follow-up into booked revenue, but the underlying idea feels similar: take something messy and give people a clearer starting point.
Would love to follow along and see how tseda develops. Also agree with Bhavya — a simple demo or walkthrough could probably help non-GitHub users understand the value faster.
@caseygaskins Thank you. I will definitely do a video explaining this. I hope that explains how you can use it with your data. I am also building a complimentary tool that explains graphs and what's happening in natural language. I hope with these two it makes sense. I am more than happy to answer questions on how to interpret a particular result.
That sounds like the right direction. A natural-language layer explaining what the graph is showing would make this much more approachable for people who are not technical enough to interpret the raw analysis on their own.
The video walkthrough will definitely help too. I think seeing one real dataset go from upload → suitability check → change point detection → plain-English explanation would make the value click quickly.
Looking forward to seeing how you build the two tools together.
This is actually pretty neat most tools stop at plotting, but tseda trying to explain the data (trend/seasonality/noise) is where real value kicks in. If you can make the insights feel obvious for non-experts too, this could be super useful beyond just data folks.
@paul__turner Thank you so much for the input. I also want to check if you had a chance to review the observation log that is automatically generated in the Observation Log panel. It might have what you are taliking about, if not can you give me some examples so that I can work on that. I alo want have another data product called KMDS (https://github.com/rajivsam/kmds) that might explain what you are talking about here. What I am working on doing is kind of getting the analyst/stakeholder involved in the tool through a NLP like interface. Basically, you provide the tool the documentation/confluence page or url and your primary interest in a particular session - it could be "understand the data" in the data dictionary. It could be implement "this algorithm" in this research paper, or, to your point "what's happening here" in this plot, you get an NLP description. The tool is really the glue between the analyst/stakeholder and the data. This feature is coming up in KMDS - next week or so. TSEDA is integrated with KMDS. Appreciate the input.