Shikhar Mishra

Ansah Technologies

• Redesigned agentic retrieval pipelines to tool-call-based ReAct architectures – improved answer accuracy to 93% accuracy.
• Built production-grade RAG systems for enterprise document intelligence (10,000+ docs) with semantic search and multi-step contextual reasoning.
• Delivered NLP search & personalization pipelines driving 2× active users & 120% longer sessions.
• Engineered scalable FastAPI microservices integrated with Apache Solr, MongoDB, MinIO, Keycloak, and LLM inference infrastructure.
• Implemented dense vector retrieval (Solr DenseVectorField, SentenceTransformers, cross-encoder rerankers) with hybrid BM25 + vector search and semantic caching.
• Designed Learning-to-Rank systems using RankLib, behavioral signals, click models, and semantic embeddings.
• Finetuned Sentence-transformer model for generating domain aware (aviation) embedding achieved 96% recall.