Ankit Singh — AI Researcher

👨‍💻 About Me I’m an applied AI Scientist with end-to-end expertise across Generative AI, NLP, Computer Vision, Graph Neural Networks, and E-commerce Re-ranking Systems. I build production-grade AI platforms that connect research to scalable deployment — from RAG / GraphRAG and multi-agent pipelines to feature-driven recommendation engines, observability, and model evaluation frameworks. -- 🧠 Core Expertise LLMs | RAG / GraphRAG | NLP | CV | GNNs | LangGraph | Multi-Agent AI | Retrieval & Re-ranking | MLOps | Model Serving | Observability | Recommendation Systems -- 🚀 Experience Highlights ✨ Generative AI & LLM Systems Fine-tuned LLaMA-3, Mistral, Phi-3, & Qwen (Full Precision / LoRA / QLoRA) for domain adaptation. Trained LLM-based rerankers via GRPO, optimizing NDCG /MRR and token-level ranking. Fine-tuned Cross Encoder/Colbert Reranker models using ModernBERT/BGE-v2-M3/BGE-M3 for boosting semantic search. Architected RAG + GraphRAG pipelines for multimodal retrieval across text, tables & visuals. Built dual-retriever chatbot (Vector DB + FHIR) for compliant clinical Q&A. Created semantic search & template recommendation engine indexing 5 M+ presentation constructs using multimodal embeddings + intent-aware ranking. Deployed Ray Serve + vLLM for distributed inference and low-latency serving. Designed multi-tenant evaluation & telemetry with Langfuse, OpenTelemetry, Grafana, LiteLLM for unified cost, latency & quality tracking (RAGAS, DeepEval). Built vendor-agnostic LLM API gateway (OpenAI, Claude, Cohere) with routing, analytics & compliance. Developed LangGraph multi-agent pipeline (Planner → Coder → Evaluator) for automated code generation to do data analysis for various downstream tasks like report generation, large excel understanding, slide generation via Claude + Aspose / Python-PPTX. --- 🎯 Ecommerce Recommendation & Re-Ranking Trained and deployed XGBoost & LightGBM models to improve template and content recommendations. Designed user-level feature store and reranking system at Walmart scale, leveraging session signals and contextual features for personalized search and reranking. --- 📷 Computer Vision & Document AI Fine-tuned YOLO, DINO, Faster RCNN for layout segmentation and multimodal understanding. Trained LayoutLM, LILT, GNNs for structure-aware reasoning & document relation extraction. --- ⚙️ Core Stack Python · LangGraph · LangChain · Ray Serve · vLLM · Bedrock Claude · Airflow · Kafka · Spark · Flink · FastAPI · Grafana · Langfuse · MLflow · Neo4j · Milvus · S3 · QLoRA · LLaMA-3 · Mistral · Phi-3

Stackforce AI infers this person is a highly skilled AI/ML engineer with expertise in Generative AI and E-commerce solutions.

Experience: 8 yrs 9 mos