Lugman Hussain Khan
Four architecture approaches for using smaller language models
Explore four architectural approaches for using SLMs to reduce inference cost and latency while preserving application quality and reliability.
Lugman Hussain Khan
Improving agentic web search in small language models
Enhancing multi-hop web search in 2-billion-parameter language models via 193 distilled trajectories.
Lugman Hussain Khan
A practical guide to LLM inference optimization
A practical guide to LLM inference optimization, covering concurrency, context length, quantization, GPU memory, vLLM vs. SGLang, and latency-throughput trade-offs.
Lugman Hussain Khan
Move beyond traditional agentic search with specialised retrieval sub-agents
Traditional agentic search bloats LLM context windows. Discover how specialised retrieval sub-agents optimise multi-hop search for long-horizon AI workloads.
Lugman Hussain Khan
Building ElectionGPT for the 2026 Tamil Nadu elections
How Madhi AI built ElectionGPT, an AI-powered election intelligence system for Tamil Nadu using grounded retrieval, production-ready architecture, and reliable political knowledge access.
Lugman Hussain Khan
Improving long-horizon planning through harness engineering
A technical exploration of improving long-horizon planning reliability through structured harness engineering, evaluation constraints, and resilient orchestration design for advanced AI systems.
Lugman Hussain Khan
Bringing visual and interactive chats for any LLM
How Weave enables browser-native visual and interactive chat experiences for any OpenAI-compatible model through streamed inline HTML rendering.