Swiggy, the Indian food‑delivery platform, announced the release of Hermes V3 on Tuesday, a generative‑AI driven assistant that lets employees ask data questions in plain English and receive SQL queries in return. The system is integrated directly into Slack, the company’s primary internal communication channel, and is designed to improve the accuracy of generated SQL and support multi‑turn analytical conversations.
Background of the Hermes Series
Hermes began as an internal project aimed at simplifying data access for Swiggy’s workforce. The first version, Hermes V1, was a prototype that demonstrated the feasibility of converting natural‑language questions into SQL statements. In 2023, Swiggy rolled out Hermes V2, which added basic conversational memory and a limited set of data sources. The new iteration, Hermes V3, builds on that foundation by incorporating advanced generative‑AI capabilities and a more robust architecture that supports complex, multi‑step queries.
Technical Features of Hermes V3
Vector Retrieval for Contextual Relevance
Hermes V3 uses vector retrieval to locate the most relevant data segments before generating a query. This approach allows the assistant to consider the semantic meaning of a user’s question, rather than relying solely on keyword matching. By embedding data into high‑dimensional vectors, the system can quickly identify related tables and columns, improving the relevance of the generated SQL.
Conversational Memory for Multi‑Turn Interaction
The tool maintains a short‑term memory of the conversation, enabling it to handle follow‑up questions that refer back to earlier parts of the dialogue. This feature is essential for analytical tasks that require iterative refinement, such as narrowing a dataset by adding filters or aggregating results across multiple dimensions.
Agentic Orchestration for Complex Query Construction
Hermes V3 employs an agentic orchestration layer that decomposes a user’s request into smaller sub‑tasks. Each sub‑task is handled by a specialized component that may involve data retrieval, transformation, or validation. The orchestrator then stitches the results together into a single, coherent SQL statement. This modular approach reduces the likelihood of errors and improves the overall reliability of the output.
Explainability for Transparency
To address concerns about black‑box AI, the assistant provides an explanation of the logic behind each generated query. Users can view the reasoning steps that led to the final SQL, which helps build trust and facilitates debugging when the results do not match expectations.
Implications for Swiggy’s Workforce
By allowing employees to query data without writing SQL, Hermes V3 is expected to lower the barrier to data access across departments. Analysts, product managers, and marketing teams can retrieve insights more quickly, potentially reducing the time required for data‑driven decision making. The system’s integration with Slack also means that users can perform these tasks within the context of their existing workflow, minimizing context switching.
Swiggy’s engineering team has indicated that the tool is currently in a pilot phase, available to a limited group of users. Feedback from this group will inform further refinements before a broader rollout. The company has not yet disclosed a public release date, but internal documentation suggests that a company‑wide deployment could occur within the next six months.
Future Outlook
Hermes V3 represents a step forward in Swiggy’s broader AI strategy, which includes initiatives in recommendation engines, fraud detection, and supply‑chain optimization. The company’s focus on generative AI for internal tooling aligns with industry trends that emphasize democratizing data access and accelerating analytics. While the tool is currently tailored to Swiggy’s internal data ecosystem, the underlying architecture could be adapted for use by other enterprises seeking to streamline data queries.
As the platform matures, Swiggy may expand Hermes V3’s capabilities to support additional data sources, such as external market feeds or third‑party APIs. The company may also explore integrating the assistant with other collaboration tools beyond Slack, broadening its reach within the organization. For now, the immediate priority remains refining the user experience, ensuring query accuracy, and gathering user feedback to guide future enhancements.
