
Databricks turned every data question from an engineering ticket into an instant answer. With Genie Agents, 30+ people self-serve insights across 200+ TB of data in plain English.
1Weather, the consumer-facing weather application by InMobi, delivers hyper-personalized weather experiences to millions of daily active users while monetizing through programmatic advertising. With 200+ TB of raw event data and no unified query layer, every business question required an engineering ticket. By consolidating data and AI on the Databricks Data + AI Platform on Google Cloud, 1Weather enabled self-service analytics with Genie Agents, unified its ad auction funnel and automated support email triage with SQL AI Functions.
Data access challenges with 200+ TB of raw event data
Delivering micro-local weather updates and running programmatic ad auctions generated massive amounts of data for 1Weather. Over 200 TB of raw event data sat as parquet files in Google Cloud Storage (GCS), with no unified query layer and no direct access for non-engineering teams. Ad auction data came from multiple demand sources, each with its own ID scheme, and extracting even a basic metric like revenue required multi-step JSON parsing with source-specific logic.
"Data prep was our biggest bottleneck," said Divya Gupta, Machine Learning Engineer 2 at 1Weather by InMobi. "Raw events were deeply nested JSON files, making cross-source auction reconciliation impossible without a lakehouse architecture."
Product managers, monetization leads and support analysts had no direct access to data. Every question meant a Jira ticket or Slack message to data engineers, who wrote ad hoc SQL and sent back CSV exports. Decisions on ad floor pricing, user engagement and customer support sat in the queue for one to five days, and context often got lost between iterations. The same questions came back every month, so the data team spent more time answering repeat requests than building anything new.
Centralizing ad tech workflows and conversational analytics with Databricks
To address these bottlenecks, 1Weather implemented the Databricks Platform to centralize event data, machine learning models and conversational analytics. The team brought governance, storage and compute together, giving both technical and non-technical users one source of truth for ad tech and business reporting.
Building a unified data architecture
1Weather uses Unity Catalog to govern 250+ tables and views across development and production catalogs. It applies the same table-level access controls whether users work in notebooks, dashboards or Genie Agents. Delta Lake with liquid clustering serves as the event store for high-throughput queries, and serverless compute is the default for interactive workloads.
Thirty streaming tables handle real-time event ingestion, while Lakeflow Jobs orchestrate daily production jobs spanning ad analytics, ML observability, location enrichment, partner reporting and cost alerts. SQL AI Functions power LLM-based customer support automation, including sentiment analysis, severity classification and auto-generated replies.
“The biggest win was having everything in one place,” explained Divya. “We run our daily ML pipelines and ad auction reconstruction on the same platform as our analytics, and that consolidation alone saved us a lot of context switching.”
Enhancing self-service capabilities with Genie Agents
To clear the ticket queue, 1Weather built a semantic layer of nine Unity Catalog views and a business definitions table. The same views power 15 active Genie Agents and 25+ AI/BI Dashboards. The monetization team, among the most active users, now asks questions such as “What’s the eCPM by ad source for interstitial ads last week?” and “Compare fill rates across demand sources” without having to file a ticket.
Accelerating developer velocity with native workspace tooling
Data engineers and data scientists share notebooks and Unity Catalog views directly in the workspace, ending back-and-forth requests to re-extract features. Practitioners use Genie Code daily for code authoring, debugging and exploratory analysis.
Eliminating data bottlenecks and enhancing ML experimentation
Before Genie Agents, the data team fielded 15 to 20 ad hoc requests each week, and each request took one to five days. Today, users across eight business domains self-serve answers on demand, with more than 3,200 questions asked to date and usage holding steady month over month. By optimizing clustering on Delta Lake, the engineering team also cut one key pipeline from over two hours to about three seconds.
Self-service analytics and unified ML operations delivered measurable impact:
- Non-technical users, about 60-65% of Genie Agent users, now self-serve insights without data team tickets.
- The customer support team’s Genie Agent has handled 850+ queries, making that workflow largely self-service.
- Genie Agents automate Play Store and support inbox replies, as well as daily and weekly stakeholder reports, freeing up about 25 hours of manual work each week.
- AI/BI Dashboard views grew about 10x and unique viewers about 5x in six months, with 10+ team members building dashboards.
- Pipelines reliably handle about 270 million daily ad auctions, with daily automated observability across 12+ concurrent floor-price experiments.
- Engineers built a REST API wrapper that programmatically queries Genie Agents, extending their capabilities beyond live Q&A.
- An alert-based system sends formatted HTML reports to product teams, ML engineering managers and model owners.
- A cost-tracking Genie Agent over system tables helps the team spot wasteful compute and manage DBU spend.
“People who had zero data access before now get answers themselves,” added Divya. “The data team builds things instead of answering repeat questions.”
Looking ahead, 1Weather plans to bring Genie Agents to sales and partnerships teams and roll out more domain-specific agents with curated semantic layers. The team will also extend its programmatic Genie Agent reporting, finish implementing its feature store architecture, replace rule-based floor pricing with ML models per ad type and explore Agent Bricks for multi-step analytical workflows.
FAQ: 1Weather by InMobi on Databricks Genie Agents
1Weather uses Genie Agents on the Databricks Data + AI Platform, enabling non-technical users across eight business domains to query 200+ TB of event data in natural language.
