I’m building NovaTrace (Starithm - Real-time Astronomical Event Intelligence Platform), a production platform that aggregates multi-messenger alerts from GCN in real time. Given the overlap with what alert brokers care about, particularly optical follow-up of GW and GRB events, I wanted to share it here and get honest feedback.
What it does right now:
Consumes live GCN Kafka streams from Fermi, LIGO/Virgo/KAGRA, IceCube, Einstein Probe, SVOM, CHIME, and DSA-110
Groups stream notices into canonical events with AI-generated summaries from notices + GCN circulars
Extracts structured measurements from circulars (RA/Dec, fluence, spectral parameters, redshift, photometry) into a queryable database
Interactive 3D sky map of recent events by position and type
NovaTrace currently doesn’t ingest ZTF/Rubin optical alerts — for GW follow-up, optical counterpart association seems like an obvious gap. Is that something useful to your workflows, or is it already well-covered by existing brokers? Boom notice type has been introduced a few days ago and has been integrated so anything new would be available, but would love to know other ways users expect the information.
Would programmatic access to event data and circular extractions be useful?
Any other streams or communities I should be talking to?
We are also working on a cross matching engine which cross matches events from different brokers to consolidate to one event and add intelligence to it.
The goal is to fill the gaps with automation to make researchers 10x more efficient.
This looks very interesting and useful, especially for cross-matching alerts across different observatories and messengers.
I am currently working with Rubin/Fink/Lasair alerts and building an experimental pipeline for moving-object and anomaly detection. One use case where your platform could be extremely valuable is connecting unusual optical alerts from Rubin with high-energy or multi-messenger events.
For example, I have an anomalous Rubin alert from May 1:
diaObjectId 170340759185129807
It is a high-reliability single-frame detection with unusual elongated morphology. When I checked your platform, I found this Fermi GRB alert: id=799400780
The time difference is only about 4–5 minutes. The sky positions are not the same, but they are not completely random either: the separation is roughly 10 degrees outside the Fermi localization radius, and the elongated Rubin anomaly appears to be oriented roughly toward the north-west, in the direction of the GRB localization region. This does not prove any physical association, of course, but it makes the pair interesting enough for manual inspection.
This is exactly where adding Rubin/ZTF optical alerts could be powerful. If NovaTrace could ingest optical alerts, identify unusual single-frame anomalies or moving-object-like detections, and then cross-match them with GRB/GW/neutrino/X-ray alerts in time and sky position, it would be amazing.
Even better would be an AI layer that can notice these links and patterns automatically and suggest follow-up.
So yes, from my perspective, Rubin/ZTF alert integration would be very useful, especially not only for confirmed counterparts, but also for anomaly discovery and follow-up recommendation.
Even better would be an AI layer that can notice these links and patterns automatically and suggest follow-up.
I think I can certainly build something like this. Where I lack is actually domain knowledge. I would love to connect with you, understand this more and build it the right way. Can I come back to you with some questions over this topic, maybe show you what I have in mind to get more of your inputs ? That would be very helpful.
I can build something that sort of “marks” an event for more manual inspection with some pre-built insights.
Thanks again for the thoughtful reply. I highly appreciate this!
Regards
Kusha
Hi Elena,
Just anted to update you before I send the email with questions.
So, I did quickly built a POC of a cross match engine with optical counterpart fetching and I tried it on GRB: 799400780 to check if it returns diaObjectId 170340759185129807. It didn’t unfortunately, the reason was that the GRB: 799400780 never got a final position alert which made its in database position (Through older alerts) with a big error circle. I tried querying Lasair with the the given GRB position but the separation is capped.
I am working on if I can get fits files for the gcn notices and use those as well to let it search across the probability region, that would take some time. I will let you know. Meanwhile if there is something that comes to your mind , do let me know.
Thanks again
You have been a huge help!
Regards
Kusha
Hi Kusha,
Thank you for checking this and for quickly building a POC cross-match engine.
I’m not sure which exact parameters your current cross-match logic uses, so I may be misunderstanding some details. But I checked Fermi trigger 799400780 against the original NASA/GCN notice, and I noticed a possible timestamp issue.
In the original NASA/GCN notice (https://gcn.gsfc.nasa.gov/other/799400780.fermi), I see the event date/time as:
GRB_DATE: 21162 TJD; 122 DOY; 26/05/02
GRB_TIME: 27975.26 SOD {07:46:15.26} UT
NOTICE_DATE: Sat 02 May 26 07:46:20 UT
So, the original notice seems to point to May 2, 2026 07:46:15 UTC. I do not see the May 1 timestamp that appears in NovaTrace as Trigger Time (T0) May 1, 2026 02:16:15 AM UTC, so I’m not sure where that time is being pulled from.
If the event timestamp is being parsed incorrectly, then this may not be a valid test case for Rubin cross-matching, and I would not expect the engine to return the Rubin diaObjectId correctly.
Also, the original Fermi notice lists this trigger as MOST_LIKELY: 100% TGF, so it may not be an astrophysical GRB candidate for optical counterpart matching.
I’m still happy to discuss this by email. I think the general idea of cross-matching Fermi events (and the others) with Rubin optical alerts is very interesting, though time consistency and event classification will be critical for making the results trustworthy.
Thank you so much for catching this, you are absolutely right. This is also what I needed, to check if the data I am showing has any inconsistencies.
The timestamp discrepancy you noticed (May 1 02:16:15 UTC vs the correct May 2 07:46:15 UTC) was a real bug in our pipeline. It was parsing the VOEvent ISOTime field (“2026-05-02T07:46:15.26”, which has no timezone indicator) as local time (IST, UTC+5:30) instead of UTC, shifting it back by 5 hours 30 minutes. I have now patched the parser and the database layer to enforce UTC, and have backfilled the existing records so you should see the correct time.
I have also added the domain knowledge of the classification [19 means TGF, etc], so the future events show it better and it gets passed to the ML pipelines for insights as well.
With the timestamp fix in place, the cross-match time window filter is now correctly anchored to the true trigger time. I will share you a google doc on email explaining my approach of crossmatching soon [just making it a bit more readable].
Yes, I’m aware of AstroCOLIBRI. it’s a great platform. NovaTrace is still very much in early phases and trying to come at it from a slightly different angle. Rather than a
visualisation-first tool, I want to build more of a structured alert intelligence layer to help researchers quickly assess what’s worth following up, structured data for research in general.
That said, there’s definitely overlap and I think platforms like AstroCOLIBRI and NovaTrace can complement each other well for the community.
I do however get curious [and wasn’t able to find much] about how widely is the AstroCOLIBRI platform used by the community ? What do they like or dislike about it ? What’s missing?
Would love to hear your thoughts and more if you get a chance to dig deeper into NovaTrace. Always happy to discuss!
I 'm a user of AstroCOLIBRI, I use the notifications via Slack (on GRBs, EP transients, bright supernovae, classified supernovae, etc.).
You can find a lot of information on AstroCOLIBRI on their Web site and Forum. They organize a workshop in Paris every year, in October. The next one has just been announced:
@iperezfournon
Hi, Ismael,
I tried Astro-COLIBRI through their web interface without registration, but I couldn’t find the Fermi alert I was interested in: https://gcn.gsfc.nasa.gov/other/799400780.fermi
Do you see this alert in your app, and are you using the installed app or the web interface?
Thank you,
Elena
I get the notifications using the mobile app. For GRBs, one can get notifications on All GRB alerts, well localized GRBs, and Burst alerts. I don’t see the one you are interested in.
In the GCN notice it is classified as TGF, Terrestial Gamma-ray Flash:
MOST_LIKELY: 100% TGF
2nd_MOST_LIKELY: 0% n/a
So, this may be the reason why it’s not in AstroCOLIBRI.
You can register in AstroCOLIBRI and the Forum and ask, or email Fabian Schuessler (fabian.schussler@cea.fr).
Just wanted to update you that I have sent you the email with a document shared explaining the approach. Let me know if you can’t access it. Looking forward to getting your insights.
Hi @OlderDyad
It doesn’t yet. But I can make that.
Tell me more about it, how the flow should look like? Exactly what input would you put it and how would the cross match work? What details would you want in the output?
I am honestly outsider to the community and have minimal background in this. But I love the domain since school and even if I didn’t get a chance to go into it, I want to utilise my skills to give back to the community, so feel free to tell me the flows and experience you would want to see in such a platform and I will make it.
Hi Kusha,
nice tool! Just a doubt: while playing with it I noticed that the position reported in the “Event details” (Overview tab) for TrigID 800881010 (https://starithm.ai/novatrace/events/800881010) is RA=0, Dec=0, Error Radius=0. It looks strange, especially the error. Maybe I’m misunderstanding something.
Regards
Giuseppe