The Nor AI engineering blog · Uttar Pradesh, India
Engineering essays. From the people who write the code.
Written by our founding engineers, not a marketing team: multi-agent orchestration, pgvector retrieval, MCP servers and serving open-weight models.
Topics
- essays
- 9
- Latest
- 20 Oct 2026
- Authors
- 3
Featured essayLLM Systems
Architecting Deterministic AI Agent Workflows for Scale
An in-depth analysis of multi-agent state transition machines, structured JSON schema validation, automated self-healing repair loops, and fault-tolerant background execution queues.
From the essay
LLM Systems
The same pipeline, end to end.
The essay starts from raw, messy input and ends in a record a machine can trust. Nothing is summarised by hand; every field is extracted, validated, and flagged.
Raw input
vllm serve model --quantization awq --gpu-memory-utilization 0.9
Verified record
- weights
- AWQ · 4-bit · INT4
- attention
- FlashAttention-2
- throughput
- measured · not estimated
- fallback
- CPU offload on OOM
Zod contract
z.object({ quantization: z.literal("awq"), attention: z.literal("fa2"), maxModelLen: z.number().positive(), }).strict()
Verified extraction, never summarised by hand.
All essays 9
- LLM Systems7 min readGourav
Eliminating Hallucinations in Enterprise Pipelines with Structured Outputs
Driving down probabilistic failure modes in mission-critical banking, legal, and HR automation with schema-enforced generation and verification circuits.
October 14, 2026 - LLM Systems7 min readSonu
Connecting Developer Tools via Model Context Protocol (MCP)
Understanding standard MCP tool servers, secure resource handlers, JSON-RPC communication, and how AI assistants interact safely with local databases and APIs.
October 10, 2026 - Infrastructure10 min readGourav
Deploying Open-Weight LLMs Locally with vLLM, AWQ & FlashAttention-2
A complete blueprint for running high-throughput, low-latency open-source models (Llama 3.3, DeepSeek, Qwen 2.5) on private infrastructure with AWQ 4-bit quantization.
October 06, 2026 - Document AI9 min readGourav
Best Practices for Hybrid Vector Search & RAG Retrieval
Key strategies for document chunking, hybrid keyword-dense embedding indexing, Reciprocal Rank Fusion (RRF), and grounded context validation in enterprise knowledge search.
October 02, 2026 - Document AI7 min readGourav
Automating Candidate Screening: Skill Extraction Patterns
Technical insights into parsing multi-format resume documents, extracting verified candidate qualifications, and computing objective match scores with auditable weighting.
September 29, 2026 - Document AI8 min readSonu
Multi-Modal Audio & Video Synthesis: Timestamped Chunking & LaTeX Math Extraction
How to turn a two-hour lecture recording into timestamped summaries, definition glossaries and clean LaTeX formula cards.
September 25, 2026 - Document AI8 min readDhruw Kumar Singh
Vernacular NLP: Engineering Hindi-English Code-Mixed Speech & Public Gazette Parsing
What it takes to turn Indian public employment gazettes, Hindi PDF tables and code-mixed announcements into fields a job seeker can trust.
September 22, 2026 - Infrastructure6 min readDhruw Kumar Singh
Operational Redundancy and Fail-Safe Engineering Principles
Applying multi-tier fallback routing, liveness probes on every long-running worker, and strict DevSecOps redundancy across high-availability background processes.
September 19, 2026
Want to publish a benchmark with Nor AI?
If you are a researcher, maintainer or engineer with a reproducible benchmark on local LLM runtimes in mind, we would like to run and publish it with you.