Delta Lake + Workflows + Unity Catalog in one platform eliminated the need for stitching together separate ingestion, transformation, and governance tools. As a data engineer, I spend more time building pipelines and less time managing infrastructure. The notebook experience and cluster auto-scaling make iterating on complex ETL fast and cost-efficient. Review collected by and hosted on G2.com.

Cluster spin-up times and cost predictability are still the biggest friction points for me. Cold starts can really slow down ad-hoc work, and DBU costs need close monitoring to avoid unpleasant surprises. The Workflows UI has improved a lot over time, but it still doesn’t feel as flexible as dedicated orchestrators when you’re dealing with more complex DAGs. Even so, I see these as mostly polish items—the platform’s core value easily outweighs them. Review collected by and hosted on G2.com.

In a telco environment handling massive data volumes from fixed and mobile networks (GPON, 4g/5g Core, and RAN) ingesting unstructured or semi-structured frequency telemetry incrementally from our virtualized functions like vEPC, vCPE or VHGW) with minimal setup.

My team works closely with virtualized network functions and Multi-access Edge Computing. Features like Streaming Tables and Materialized Views help us to build low-latency pipelines that process network performance metrics near real-time, helping us monitor network KPIs and QoS efficiency.

Because my team's core experties lies in network deisgn and system virtualization rather than database administration, Predictive Opimization and Liquid Clustering are highly beneficial. Tehy autonomously handle table maintenance, file compaction, and data layout optimization freeing up our resources to focus on network architecture. Review collected by and hosted on G2.com.

Virtualized network functions, routers, and disaggregated hardware frequently undergo software updagrades, which often introduce sublte changes in telemetry output schemas. When using structured streaming or auto loader these schema drifts cause our streaming queries to fail, requiring a manual restart of the stream to re-plan the schema.

When we need to update the logic of a complex network KPI defined within a materialized view, any change to the query triggers a full recomputation of the view. Given the massive scale of telecom transaction datasets, this can result in noticeable compute costs.

We rely on a variety of data tools within our ICT ecosystem, not all solutions featured in Partner Connect natively support Unity Catalog. This can crete integration and governance hurdles when we try to connect certain third-party analytics and data preperation tools to our secured data lake. Review collected by and hosted on G2.com.

The ecosystem. What I like most about Databricks is how it removes a lot of the usual mess you run into with data work. Instead of juggling separate tools for engineering, analytics, and ML—and then spending extra time getting them to talk to each other—it brings everything into one place. That alone cuts down a lot of friction and saves time.

I also like the Lakehouse idea because it feels genuinely practical: you don’t have to choose between a data lake and a warehouse. You can work with one unified setup and still get performance when you need it.

On a day-to-day level, it’s also nice that different teams can collaborate in the same environment without constantly copying data around or rebuilding pipelines. Overall, it keeps things simpler and faster, especially when you’re iterating. Review collected by and hosted on G2.com.

What I don’t like about Databricks is that it can feel a bit heavy when you’re just trying to do something simple. There’s a lot going on under the hood, and while that’s great for scaling, it also comes with a learning curve. Things like clusters, configurations, and job setup take some time to get comfortable with.

Cost is another concern. Usage can creep up quickly if you’re not actively monitoring it, especially when teams can spin up compute freely. And at times, the overall experience feels a little fragmented across notebooks, jobs, and repos, rather than being one smooth, unified flow.

So, yes—it’s powerful, but it definitely takes discipline to keep things clean, efficient, and under control. Review collected by and hosted on G2.com.

What I like most about Databricks is how it centralizes data engineering, analytics, and AI in a single platform, which greatly facilitates the workflow on a daily basis. The integration between notebooks, pipelines, and distributed processing makes development faster and more organized, especially in projects with a large volume of data and automations.

Another point that I consider very strong is the experience with Apache Spark, integrated in a simplified way. Even in more complex scenarios, the performance is usually excellent, allowing large-scale data processing with good stability and scalability. This greatly helps in integrations, ETLs, and analyses that, in other solutions, would require much more effort. Review collected by and hosted on G2.com.

Although I quite like the platform, some aspects of Databricks can still be challenging. The main one is the cost, especially in environments with intensive processing or when clusters are not well optimized. Without more rigorous usage control, expenses can increase rapidly.

Another aspect is the learning curve, which can be steep for teams that are starting in the distributed data ecosystem. Concepts related to Spark, clusters, optimization, and resource management require time to adapt, especially for those coming from more traditional tools.

In UI/UX, although the interface is generally good, some administrative processes and more advanced configurations can seem confusing at first. In certain scenarios, identifying performance or permission issues may also require more technical knowledge. Review collected by and hosted on G2.com.

I really appreciate how Databricks earns its place in my work as the analytics tier for the reference architectures I propose to customers. It simplifies the process by allowing telemetry from distributed device fleets to land directly, enabling me to model aggregation, anomaly detection, and historical comparison without needing separate infrastructure. The platform's cost efficiency is a decisive factor for me, and it delivers more value than its price suggests. Retrieval speed is impressive, letting me query a warehouse live during design sessions without relying on others to run the request. The setup required far less scaffolding than anticipated, positively impacting how I prototype. Starting with Databricks didn't involve the tedious configuration I expected from a platform of its scale, so the environment was productive almost immediately. Also, the platform significantly shortens the path from an idea to a prototype that clients can see, thanks to its low cost, quick retrieval, and minimal setup requirements. Review collected by and hosted on G2.com.

Handling of unstructured content remains the weak point in my view, since building a usable search layer over message archives, PDF documents and finance spreadsheets proved awkward enough that I compromised on the design. Vector search performance disappoints as well: direct lookups return promptly, yet reasoning queries slow noticeably, which becomes apparent the moment a client is watching a demonstration. Review collected by and hosted on G2.com.

I really appreciate Databricks' ability to consolidate scattered infrastructure data like permitting records and contractor reporting into a single workspace, which makes spotting compliance gaps and making strategic decisions much easier. The broad data source connectivity is also a highlight for me because it bridges legacy systems with much less friction than I expected. Having Genie to pair with it has been a game changer, allowing me to build, test, and deploy data objects quickly, keeping my analysis moving without stalling. The initial setup was surprisingly simple, taking just a few hours with no major problems. These features make Databricks valuable for someone like me who needs reliable data more than deep engineering expertise. Review collected by and hosted on G2.com.

My one reservation concerns the Python editor. It serves its purpose, yet it could stand to be more robust. A richer editing experience would make the occasional bit of custom scripting feel less cumbersome when I am refining a data object. I'd like to see richer code intelligence in the Python editor, things like smarter autocomplete, inline error highlighting, and better debugging tools, so that when I write custom scripts to shape a data object, the process feels smoother and I catch mistakes before running the code rather than after. Review collected by and hosted on G2.com.

I really like the power of its engine, since it’s based on Apache Spark, especially when processing large volumes of telemetry, diagnostic logs, or performance metrics from embedded devices during stress tests. I also appreciate the support for collaborative notebooks in Python and Scala, which makes data exploration, building fast ETL pipelines, and running models much easier. The integration with Unity Catalog for governance is good as well; I like it. Review collected by and hosted on G2.com.

I don’t like the cost model because if I’m not careful with the cluster size or query optimization, the bill at the end of the month can be quite a surprise. Review collected by and hosted on G2.com.

What I like best about Databricks is how it simplifies large-scale data processing and collaboration in one platform. The integration with Spark and cloud service makes handling big data much more efficient. I also like the notebook environment, which makes it easier for teams for works together on analytics and machine learning tasks. Review collected by and hosted on G2.com.

One thing I dislike about Databricks is the platform can feel complex for new users, especially when managing clusters and configurations. Pricing can also become expensive with larger workloads if resources are not optimized carefully. While integrations and AI features are powerful, the onboarding process and support documentation could be more beginner-friendly. Review collected by and hosted on G2.com.

The UX is one of the strongest parts. The notebook experience is clean and intuitive, collaboration is straightforward, and moving between exploration, experimentation, and production workflows feels seamless. It has enough flexibility for advanced users while still being approachable enough that onboarding new team members is fast. People can usually become productive quickly without spending weeks learning platform-specific quirks.

The integrations are also excellent. It works smoothly with the broader cloud ecosystem and connects well with data sources, orchestration tools, model serving infrastructure, and external systems. That interoperability makes it much easier to move from prototype to deployed pipeline without constantly rebuilding connectors or managing glue code.

Performance has been consistently strong, especially when working with distributed workloads and large-scale feature engineering. Spark optimization, cluster management, and managed infrastructure significantly reduce operational overhead, which lets me focus more on model development and analysis rather than environment tuning. For iterative experimentation, spin-up times and overall responsiveness are noticeably better than many alternative managed platforms. Review collected by and hosted on G2.com.

One area where Databricks could improve is pricing. The platform delivers strong capabilities, but costs can escalate quickly for high-frequency or real-time workloads. For use cases involving continuously running low-latency tick pipelines, streaming market data, or iterative model retraining, the pricing can become fairly steep relative to the infrastructure being consumed. It sometimes feels like there’s a meaningful premium for convenience and managed orchestration, which can make cost optimization a constant consideration.

The AI integration is another area that still feels somewhat uneven. While there’s a clear push toward positioning the platform as an end-to-end AI/ML environment, some of the newer AI-focused features feel more like ecosystem additions than deeply integrated workflow improvements. In practice, there are still cases where custom tooling or external frameworks provide more flexibility and transparency, particularly for specialized model development, experimentation, and real-time inference use cases.

There can also be some complexity around tuning clusters and managing costs efficiently at scale. While the abstractions are helpful, getting the best performance-to-cost ratio sometimes requires deeper platform knowledge than the “fully managed” positioning might imply.

Overall, the platform is very strong technically, but pricing for always-on data-intensive workloads and the maturity of some AI-native capabilities are the two biggest areas where I’d like to see improvement. Review collected by and hosted on G2.com.

Genie code and the inline Assistant were the most helpful tools for me on my project. They helped me debug a 2k-line codebase and clearly explained why I wasn’t getting accurate data. It also provided a query to run in my source system (SQLMI). By running the discrepancy script in parallel on the source and target, I was able to debug the entire code much faster and improve my productivity. Overall, it cut my work time from about 8 hours down to around 1 hour. Review collected by and hosted on G2.com.

In Delta Sharing, there isn’t a catalog-level SELECT permission, and I sometimes think having that would be helpful. Also, when I use the Genie code inside a VM, it can make the website unresponsive at times. These are areas that could be improved. Review collected by and hosted on G2.com.