A leading American multinational confectionery company set out to modernize SnackGPT, transforming it from a prototype into a production grade Food Safety Data Intelligence platform capable of supporting approximately 90 internal and external partners by 2025. The vision required enterprise level accuracy, low latency, hallucination mitigation, and strict compliance with responsible AI standards—critical for a food safety–focused use case.
The existing implementation lacked the robustness needed for large scale adoption. It required significant improvements in factual accuracy, reliability, observability, and model governance to support thousands of queries generated by diverse user groups across global operations. Ensuring safety, stability, and compliance was paramount.
Infosys partnered with the client to re architect and productionize SnackGPT using Google Cloud’s Vertex AI ecosystem. The modernization introduced Retrieval Augmented Generation (RAG), advanced guardrails, continuous evaluation, and enterprise ready operational controls. The solution was engineered to deliver consistent accuracy, predictable performance, and scalable adoption across development, QA, and pre production environments.
Beyond modernizing SnackGPT, Infosys established a repeatable, production ready GenAI blueprint for future deployments—combining responsible AI principles, measurable business impact, and deep alignment with Google Cloud’s GenAI capabilities, strengthening its candidacy for the Google Partner Award.
90%+
Factual accuracy
<5%
Hallucination rate
3-5sec
Query latency
75%+
Active user adoption
6,000+
Distinct queries
Key Challenges
- Low maturity of GenAI prototype for enterprise adoption
- High risk of hallucinations in food safety scenarios
- Latency and stability issues under multi user load
- Lack of observability, quality tracking, and governance
- Responsible AI compliance requirements for global operations
Ready to experience?
TALK TO EXPERTS
Infosys Approach
- Re architected SnackGPT using Vertex AI Gemini and RAG patterns
- Ingested critical data sources into an analytics ready BigQuery environment
- Engineered prompts and workflows to reduce ambiguity and improve accuracy
- Implemented guardrails using Responsible AI Toolkit and iterative testing
- Established Dev, QA, and Pre Prod LLM environments
- Built continuous feedback and quality evaluation loops
- Designed Product Scorecards to track performance, usage, and adoption
The Solution
Production grade GenAI platform enabling safe, accurate, and scalable food safety intelligence
Infosys modernized SnackGPT by building a production ready GenAI platform on Google Cloud, leveraging Vertex AI Gemini for large language model responses and Vertex AI Vector Search for Retrieval Augmented Generation. Structured and unstructured data sources were ingested into BigQuery to enable analytics ready, low latency query execution.
Advanced guardrails and hallucination mitigation strategies were implemented through customer data fine tuning, iterative validation, and Google’s Responsible AI Toolkit. The solution introduced comprehensive observability using Cloud Logging and Monitoring, along with a Product Scorecard to measure accuracy, latency, throughput, and user adoption.
Continuous improvement loops integrated user feedback, automated quality evaluation, and performance dashboards. Infosys also delivered a full LLM lifecycle environment across Dev, QA, and Pre Production, ensuring stability and scalability. The result is a secure, compliant, and repeatable GenAI platform designed for real world food safety intelligence at enterprise scale.
Business Outcomes
Achieved 90%+ factual accuracy for validated “What” queries
Reduced hallucinations to below 5% with no critical violations
Delivered 3–5 second latency for 90% of pilot
Enabled strong adoption with 75%+ active pilot users
Supported 6,000+ distinct queries across pilot environments
Demonstrated consistent improvements across Product Scorecard metrics
Benefits
Modernized GenAI platform delivers accuracy, speed, safety, and scalable adoption for food safety intelligence.
- High factual accuracy for safety critical food intelligence queries
- Predictable low latency performance across environments
- Reduced hallucination risk with responsible AI guardrails
- Strong user adoption and engagement
- End to end observability and quality measurement
- Scalable, repeatable blueprint for future GenAI initiatives
