# Bharat Singh > Personal technical publication on Information Retrieval (IR), Neural Search Architecture, RAG at Scale, and Production Multi-Agent Systems. > Authored by Bharat Singh, Tech Lead for the AI vertical at Milestone Internet. ## About the Author Bharat Singh is an AI Systems Architect and Engineering Lead with 20+ years of experience directing high-throughput Information Retrieval (IR) architectures, autonomous agent platforms, distributed systems, and knowledge graphs across European institutions and enterprise SaaS. His research background centers on semantic data modeling and entity extraction across complex scientific corpora, including contributing to the research paper "The value of data" in **Nature Genetics** (2011, doi:10.1038/ng0411-281) alongside peer-reviewed publications in **JAMIA** and the **Journal of Biomedical Informatics**. Currently, he serves as Tech Lead for the AI vertical at Milestone Internet, where he directs the architecture of enterprise autonomous multi-agent workflows, neural retrieval pipelines (Late Chunking, hybrid search, cross-encoder rerankers), and core CMS engine modernizations scaling across 3,500+ commercial websites under sub-50ms p95 retrieval latency SLAs. ## Core Technical Publications - [Building Self-Discovering AI Agents: Dynamic Tool Discovery with gRPC Reflection and LangGraph](https://bsbhadwal.com/posts/architecting-dynamic-self-discovering-ai-agents-a-langgraph-and-grpc-reflection-prototype/): Architecting self-discovering AI agents using gRPC Server Reflection and LangGraph for dynamic schema discovery without hardcoded API tools. - [ColBERT and Late Interaction at Scale: Why Multi-Vector Retrieval Strains Production Memory](https://bsbhadwal.com/posts/colbert-late-interaction-at-scale-memory-bandwidth-physics/): Deep architectural analysis of ColBERT late interaction multi-vector retrieval, memory bandwidth physics, PLAID indexing, and sub-50ms latency SLAs. - [Context Caching vs. Semantic Caching: Latency Trade-Offs Across High-Throughput Agent Workflows](https://bsbhadwal.com/posts/context-caching-vs-semantic-caching-latency-physics/): Architectural comparison of LLM context caching and semantic caching. Analyzing TTFT, token economics, cache hit ratios, and multi-agent latency budgets. - [Deterministic Guardrails in Stateful Multi-Agent Loops: Preventing Infinite Trajectory Drift](https://bsbhadwal.com/posts/deterministic-guardrails-stateful-multi-agent-loops/): Engineering deterministic state-machine guardrails in autonomous multi-agent loops to prevent trajectory drift, infinite loops, and token budget exhaustion. - [Evaluating RAG Without Ground Truth: Production Pitfalls of LLM-as-a-Judge](https://bsbhadwal.com/posts/evaluating-rag-without-ground-truth-llm-as-a-judge-pitfalls/): Production analysis of evaluating enterprise RAG pipelines without human ground truth, mitigating LLM-as-a-judge bias, and building empirical synthetic test sets. - [The Paradigm Shift to Late Chunking in Retrieval-Augmented Generation: A Technical and Comparative Analysis](https://bsbhadwal.com/posts/late-chunking-in-retrieval-augmented-generation/): Technical breakdown of Late Chunking in RAG architectures. Contrasting Jina AI token mean pooling and Voyage AI contextualized embeddings against naive chunking. - [Reciprocal Rank Fusion vs. Cross-Encoders: Designing the Sub-50ms Reranking Funnel](https://bsbhadwal.com/posts/reciprocal-rank-fusion-vs-cross-encoders-reranking-funnel/): Designing high-throughput hybrid retrieval reranking funnels combining Reciprocal Rank Fusion (RRF) and Cross-Encoder neural scoring under sub-50ms SLAs. - [From Semantic Nanopublications to Knowledge Graphs: What 15 Years of Machine-Readable Data Taught Us About RAG Grounding](https://bsbhadwal.com/posts/semantic-nanopublications-to-knowledge-graphs-rag-grounding/): Applying 15 years of semantic web nanopublication principles and RDF graph triples to establish verifiable provenance and mitigate hallucination in enterprise RAG pipelines. - [Shrinking Billion-Scale Vector Indexes: Combining Matryoshka Embeddings with Binary Quantization](https://bsbhadwal.com/posts/shrinking-billion-scale-vector-indexes-matryoshka-embeddings-binary-quantization/): Techniques for shrinking billion-scale vector indexes using Matryoshka Representation Learning (MRL) and 1-bit binary quantization with Hamming distance rescoring. - [The Scaling Dilemma: An Analysis of Tool Quantity and its Impact on ReAct Agent Performance](https://bsbhadwal.com/posts/the-react-scaling-dilemma/): Comparative architectural analysis of how expanding tool quantities degrades ReAct agent reasoning accuracy, burns latency budgets, and necessitates hierarchical routing. - [The Retrieval Fallacy: Why Massive Context Windows Don't Eliminate the Need for IR](https://bsbhadwal.com/posts/the-retrieval-fallacy-why-more-tokens-dont-equal-better-context/): Why 1M+ token context windows do not replace Information Retrieval. Examining needle-in-a-haystack decay, attention dilution, and cost economics. - [The Unit Economics of Inference: When to Kill Third-Party AI APIs](https://bsbhadwal.com/posts/the-unit-economics-of-hosted-inference-vs-apis/): Financial and latency model analyzing the tipping point for migrating from third-party LLM APIs to self-hosted vLLM/TGI GPU clusters at enterprise scale. - [The Cognitive Tax of Trivial Helpers: Lessons in Interface Design from the Linux Kernel](https://bsbhadwal.com/posts/think-for-yourself-lessons-from-linus/): Analyzing Linus Torvalds rejection of trivial helper abstractions in the Linux kernel and its lessons for clean software and AI interface design. - [Why Workflow DAGs Break at 2,500+ Tenants: Moving Multi-Tenant RAG from Batch Orchestration to Event-Driven Pub/Sub](https://bsbhadwal.com/posts/why-workflow-dags-break-at-scale-pubsub-vs-prefect/): Why static workflow DAG orchestrators collapse under multi-tenant RAG workloads at scale, and how event-driven Pub/Sub message buses solve tenant isolation. ## Biographical & Professional Links - [About Bharat Singh](https://bsbhadwal.com/about/): Background, research history, technical leadership focus, and direct contact. - [Curriculum Vitae (HTML)](https://bsbhadwal.com/resume/): Complete 20+ year curriculum vitae with sub-50ms p95 SLA metrics. - [Download Curriculum Vitae (PDF)](https://bsbhadwal.com/resume.pdf): Official 2-page compiled curriculum vitae. - [GitHub Profile](https://github.com/bsbhadwal): Open-source repositories and code prototypes. - [RSS Syndication Feed](https://bsbhadwal.com/index.xml): Syndication feed of all architecture notes. ## Machine-Readable Corpora - [Full Text Corpus](https://bsbhadwal.com/llms-full.txt): Complete concatenated plain-markdown text of all technical publications for zero-latency LLM context ingestion.