Swiggy Rolls Out Hermes V3: From Text-to-SQL to Conversational AI

2026-01-02 · Source: InfoQ · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Intermediate, quick

Summary

Swiggy has launched Hermes V3, a GenAI-powered text-to-SQL assistant integrated into Slack, allowing employees to query data using natural language. This iteration significantly improves upon earlier versions that struggled with derived metrics, conversational context, and result consistency. Hermes V3 achieves this by incorporating a vector-based prompt retrieval system built on historical SQL queries in Snowflake, which converts SQL into natural language explanations for few-shot examples. It also features conversational memory for multi-turn interactions, an orchestrator agent using a ReAct-style reasoning loop to break down complex questions, and an explanation layer that provides query assumptions and confidence scores. The system is integrated with Swiggy's security, compliance, and metadata infrastructure, utilizing role-based access control and audit logs.

Key takeaway

For AI Architects and NLP Engineers building internal data querying tools, Hermes V3 demonstrates how combining vector retrieval, conversational memory, and agentic orchestration can dramatically improve text-to-SQL accuracy and user trust. You should consider reconstructing query intent from historical SQL to enhance few-shot learning and implement an explanation layer to surface assumptions and confidence scores, thereby increasing adoption among non-technical stakeholders.

Key insights

Hermes V3 enhances text-to-SQL with vector retrieval, conversational memory, and agentic orchestration for improved accuracy and user trust.

Principles

• Reconstruct missing query intent from historical SQL.
• Ground new requests in prior analytical patterns.
• Provide transparency for machine-generated insights.

Method

Hermes V3 uses vector-based prompt retrieval from historical SQL, maintains conversational memory, and employs a ReAct-style orchestrator agent for intent parsing, metadata lookup, and SQL generation.

In practice

• Convert SQL queries to natural language for vector indexing.
• Implement ReAct-style agents for complex query breakdown.
• Integrate role-based access control for data governance.

Topics

• Text-to-SQL
• Conversational AI
• Generative AI
• Vector Databases
• Few-shot Learning

Best for: AI Architect, NLP Engineer, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• AI Agent Amnesia? Here’s the Open-Source Fix That Works. — Hermes Agent uses SQLite FTS5 and LLM summarization for persistent, cross-session AI memory.
• DecoSearch: Complexity-Aware Routing and Plan-Level Repair for Text-to-SQL — DecoSearch improves Text-to-SQL accuracy for complex queries via complexity-aware routing and plan-level repair.
• Quality Over Clicks: Iterative Reinforcement Learning for Early-Stage E-Commerce Query Suggestion — Cold-EQS uses intrinsic quality rewards and uncertainty sampling to enable effective e-commerce query suggestion in cold-start scenarios.
• BADGER: Bridging Agentic and Deterministic Evaluation for Generative Enterprise Reasoning — BADGER unifies text-to-SQL and agentic reasoning evaluation for enterprise AI, validated by human expert judgment.
• We Evaluated ChatGPT vs. Google on 500 Search Queries — ChatGPT significantly outperforms Google on coding queries and ties it on general informational queries, despite not being optimized for search.
• ConvMemory v3: A Validity Context Layer for Conversational Memory via Target-Conditioned Relation Verification — ConvMemory v3 enhances conversational memory by verifying memory validity through target-conditioned relation verification.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by InfoQ.