Comet for SageMaker Partner AI Apps

I mainly use Comet  for research topics, summarizing information, and understanding difficult concepts. I use it for organizing and tracking my work on academic projects. It helps me keep track of experiments, compare results, manage data, and document my progress in one place. As a student, this makes it much easier to stay organized, analyze outcomes, and collaborate with classmates when working on research or machine learning projects.

Recently, I used Comet  while working on a machine learning project that predicts student academic performance based on study habits and attendance data. I tracked different model runs, recorded parameters and results, and compared performance metrics such as accuracy and precision. Using Comet made it much easier to identify which model performed best and keep all my experiment details organized throughout the project.

Comet helps me maintain a clear record of my work, which is especially valuable in balancing multiple assignments and projects. Instead of manually tracking results in different files, I can keep experiments, metrics, and notes organized in one place. This improves reproducibility, makes it easier to revisit previous work, and saves time when preparing reports or presentations.

The features that stand out most to me are experiment tracking, performance visualization, and project organization. Experiment tracking makes it easier to compare different models, runs, and understand what changes led to better results. The visualization tools help me quickly analyze metrics and spot trends without having to create charts manually. I also appreciate how Comet keeps datasets, code versions, notes, and results organized in one place, which makes managing projects much more efficient.

The feature I rely on the most is experiment tracking. When I am testing different models or configurations, it is incredibly helpful to have all the parameters, metrics, and results automatically logged and organized. It saves me from manually documenting everything and makes comparisons much easier. As for specific tools, I use the experiment comparison dashboard all the time. Being able to view multiple runs side-by-side and quickly compare metrics such as accuracy, loss, and validation performance helps me make decisions much faster.

Comet does an excellent job of bringing different parts of the workflow together in one platform. Instead of switching between spreadsheets, notebooks, and separate tracking tools, I can see experiment metrics, visualizations, and notes in a single place. This not only saves time but also makes collaboration and project reviews much easier.

My experience with Comet has been very positive, but there are a few areas where it could be improved. One area is the learning curve for new users. Some of the more advanced features can feel overwhelming at first, especially for students who are new to machine learning experiment tracking. More beginner-friendly tutorials and guided onboarding would help. I would also like to see more customization options for dashboards and visualizations, making it easier to create views tailored to specific projects. Another improvement would be deeper integration with commonly used collaboration tools, which would streamline project documentation and team workflows.

There are a few additional areas where Comet could improve. From a performance perspective, I occasionally notice that dashboards with a large number of experiments can take longer to load or navigate. Regarding documentation, while the available resources are helpful, I would appreciate more beginner-focused examples, step-by-step tutorials, and real-world use cases. For support, my experience has generally been good, but having more community resources, discussion forums, webinars, or educational content specifically aimed at students and researchers would be valuable.

I have been using Comet for approximately eight months.

Comet has been generally stable and reliable.

In my experiments, Comet has handled scalability reasonably well for the types of projects I work on. For moderate increases in workload, such as more hyperparameter sweeps or additional experiment runs, it still performs well and keeps the data organized in a way that is easy to navigate and compare. That said, when the number of experiments grows significantly, I have noticed that loading dashboards and browsing through large experiment histories can become slower. It is not a blocker, but it does highlight that performance can vary depending on project size. Overall, I would say Comet scales very well for academic to mid-sized machine learning projects, and it remains usable.

Customer support is pretty good, but I have not had a chance to directly reach out to them because I was able to troubleshoot all the issues with the online discussion forums. However, I heard from my colleagues and friends that customer support is actually good.

I mainly relied on a combination of manual tracking methods, such as Jupyter notebooks, Excel, or Google Sheets. I switched to Comet because it brought all of these pieces together into a single platform. The main reason for the switch was efficiency and reproducibility.

Before choosing Comet, I explored TensorBoard, Weights & Biases, and setup using Jupyter notebook spreadsheets, which is what I initially started with. I did not do a formal head-to-head evaluation, but I explored them enough to understand their workflows. I chose Comet because I felt it had a good balance of ease of use and clean visualization tools without being too complex for my projects.

I do not calculate ROI in financial terms, but I have seen it in terms of time saved, productivity, and experiment efficiency. I estimate I spend around thirty to forty percent less time organizing and comparing experiment results compared to manual tracking. Project iteration cycles are faster, and I complete research projects more efficiently. In terms of qualitative ROI, the biggest benefit is improved workflow structure and reproducibility.

Most of the major improvements I would like to see have already been covered, but one would be enhanced collaboration features.

I would suggest setting up Comet properly from the start and using it consistently for every experiment, even small ones. I also recommend taking time early on to learn how experiment tracking, metrics logging, and comparison views work because those are the features that provide the most value once you are actually iterating on models. Another recommendation is to keep experiments well-organized with clear naming conventions and tags.

I would rate my overall experience with Comet an 8 out of 10.

My main use case for Comet  is looking at investigations related to health and safety issues and incidents. For a health and safety investigation, I create a timeline of the events that led up to the incident and use Comet 's features to see whether some of these events could have been avoided.

The best features Comet offers include the ability to create the timeline in as much detail as possible and to share it with others.

I usually share the timeline with others via a link, but sometimes via export.

Comet has positively impacted my organization because it's a good system, and the interface is really simple and easy to use, which allows everyone to have a look at these health and safety investigations, of course, if they have access through the company itself. It's a good way to share knowledge amongst people.

For example, with the health and safety incidents, being able to share the timelines with others allows them to distribute it to their sites and create toolbox talks to help those on site understand the situation that's happened and how they can avoid it happening at their site.

I'm not sure how Comet can be improved, as I've only been working on it for three months, so I feel like I need a bit more time to be able to answer this question.

I have been using Comet for three months.

Comet is stable in my experience.

The scalability of Comet is fine, as you can do either one investigation or multiple, which is in regards to the work that myself and my team do.

I'm not sure how Comet's customer support is, as I'm yet to use it. From my three months of usage so far, I would say Comet is a good product to use, and if others have any issues, they should contact the customer support. It's a very good interface, simple to use.

I did not use a different solution before Comet.

I haven't seen a return on investment with Comet, not that I'm aware of, as I wouldn't have those figures being a graduate, but it may potentially have saved the team some time from whatever their previous method was.

I wasn't part of the evaluation process before choosing Comet, as I had joined, and Comet was already the chosen product.

On a scale of 1 to 10, I would rate Comet overall a nine. I chose nine out of 10 because so far I haven't run into any issues regarding the Comet interface, and I've been able to do my job well using it.

My experience with pricing, setup cost, and licensing wasn't something that I experienced directly, as it was done through the company. They had given me the laptop, and it already had access, so I just needed to confirm it with someone. I believe the company has a license, and you just have to message the licensing team to have access.

Comet is deployed in my organization as I believe on-premises, so only the laptops associated with the company can use Comet. My overall review rating for Comet is 9 out of 10.

Comet  serves as my end-to-end AI observability platform, where I integrate MLOps and AI with machine learning development. I use Comet  in my workflow for tracking purposes, where I monitor code changes during training runs, and I utilize it for model registry and storage.

The standout features of Comet that I find especially useful include LLM-specific evaluations, tracing, and debugging, along with easy deployment abilities on cloud and self-hosted on-premise solutions. I utilize it in a hosted environment at my end. Additionally, features such as integrations with frameworks like PyTorch , TensorFlow , Hugging Face , and LangChain significantly aid in building enterprise-grade applications while maintaining data sovereignty.

I have built PyTorch  programs and leveraged libraries inside some of my POCs and integrated them with Comet, which helps save time and enables me to utilize my already tested features. Comet has positively impacted my organization by facilitating the integration of my AI implementation, which saved research time and enhanced integration with existing frameworks, allowing me to leverage my existing code and libraries. The ability to debug and conduct what-if analysis across new experiments enabled me to run programs on Comet quickly and receive feedback, refining the entire feedback loop and saving time on new research and adaptations to developments in AI and generative AI.

I save approximately thirty percent of overall time in the release cycle thanks to Comet.

Some areas I believe Comet can improve include scalability limits, as I face challenges when scaling. Enhancing UI customization would leverage themes within my organization, and expanding on quant trading-specific features would be beneficial, especially since I am focusing on algorithmic trading features and mathematical model enhancements. Scalability, UI and visualization enhancements, as well as including more mathematical models, would be improvements I would appreciate.

I have been using Comet for one year.

My advice for others looking into using Comet is to leverage the integration features, as they allow you to quickly utilize existing frameworks, libraries, and code from various areas. This is one of its key features. By leveraging that, engaging in small project POCs can help you discover related experimental data, signal detections, or any mathematical models, thereby saving considerable time in research. I would rate this product a nine out of ten.

I use Comet  for summarizing articles and videos and getting PDFs instantly to draft emails and plan trips. I extract insights, which is the primary function I use Comet  for most of the time.

My end use involves automation plus agent behavior, so it can interact with websites for me, execute multi-step workflows, search, and compare between them, then act upon my instructions. This is what I appreciate the most about it.

The best features Comet offers include the agentic capability that I previously explained, where it compares and acts upon my instructions, goes through websites, makes summaries, and drafts emails, which is what I actually appreciate most in Comet browser.

Comet has made me faster in going through each article, which may not seem useful until I read the complete article. Using the summarization feature in Comet allows me to read the summary and know whether it is the relevant article that I want to look at.

I estimate that around ten to fifteen percent of my time might be saved using Comet, though the improvement is not substantial.

The agent technology hallucinates frequently, so it can give me wrong summaries or decisions and misinterpret some information. I believe it is not fully developed; however, for small tasks such as drafting, it performs adequately.

Automation is something I still need to explore more fully to understand the complete automation features of Comet.

Comet can improve by decreasing hallucinations and addressing security issues. There are vulnerabilities to prompt injection attacks, and the AI can be tricked into leaking data or acting harmfully. Improvements in security and applying regular patches could help significantly.

The user experience is acceptable, but a more modern look would enhance it.

I have been using Comet since its first release.

Comet is fairly stable, though I am not entirely certain about its complete stability.

I believe Comet can handle more users or larger workloads if needed.

I have not reached out to customer support at any time.

The time I saved is around ten to fifteen percent compared to what I have done in traditional browsers. While that is not a significant improvement, it has helped me with summarizing and drafting emails.

My experience with pricing, setup cost, and licensing is that it was all free.

I choose a rating of six out of ten for Comet because it is not fully developed. I recognize it might be the first release and the first version of what they are building, so I expect more improvements in the future.

I recommend Comet to those who are learning, conducting research, or are college students and university graduates who want to read through lengthy articles. My overall rating for this product is six out of ten.