Ganeshkumar Istharla

BioNiDi

• Designed and implemented an end-to-end Retrieval Augmented Generation pipeline for more accurate response generation based on context.
• Designed and implemented a Retrieval Augmented Generation (RAG) based application for PDF document retrieval and automatic text summarization.
• Designed and implemented a vector database for embedding storage and document retrieval based on user queries.
• Developed backend modules using Fast API and REST APIs for authentication, content management, and user workflows.
• Built frontend components using React and Next.js, integrating with backend APIs.
• Contributed to CI/CD workflows with GitHub Actions for automated testing and deployment, enhancing integration efficiency.

Ora-Share

• Architected enterprise-scale RAG ingestion pipelines using OpenAI Vector Stores, implementing automated document chunking, semantic indexing, and permission-trimmed retrieval to enforce 100% tenant data isolation across multi-tenant deployments.
• Engineered low-latency streaming AI chat experiences using Vercel AI SDK (streamText) and React Hooks, integrating custom citation mapping that generated interactive page-level references for retrieved enterprise documents.
• Built structured LLM generation pipelines using generateObject and Zod schema validation to automatically classify unstructured documents and generate hierarchical folder structures with validated outputs.
• Designed dynamic multi-model routing across OpenAI GPT models and Google Vertex AI Gemini, optimizing model selection based on latency, response quality, and token cost for diverse enterprise workloads
• Developed resilient AI orchestration layers with automated retry logic, exponential backoff, provider failover, and quota-aware error handling, ensuring 99.9% application availability during production traffic.
• Implemented prompt engineering strategies, structured system prompts, and evaluation workflows that improved response consistency and reduced hallucinations in enterprise knowledge retrieval.
• Built AI observability pipelines for monitoring latency, token consumption, retrieval quality, and model performance, enabling continuous optimization of production GenAI applications