50x cheaper GPU, 5x $/token. Why? Qwen3.8 27B vs DeepSeek V4 Flash10x bigger DeepSeek V4 is 284 Billion parameters. Qwen3.8 is 27 Billion parameters. 50x expensive hardware Qwen3.8 27B FP8 fits on a single RTX 6000 Blackwell (approx 15000 USD - 15 Lakh INR). DeepSeeOct 11 · 3 min readtechaarvam.hashnode.dev
Our in-house agent router: Jev vs Kev-4B vs bge-smallIn our (currently) in-house business automation agent, we route prompts to the right agent. This is done using an agent router. Take-away: before a prompt reaches the active agent, two small local modOct 10 · 3 min readtechaarvam.hashnode.dev
Puzzle Solution Revealed - Transformer: Need for Position EmbeddingIntroduce position embedding, residuals and layers ▶ Watch the 7-minute walkthrough import torch torch.set_printoptions(precision=2, sci_mode=False, linewidth=160) In this notebook we are extendingSep 15 · 14 min readtechaarvam.hashnode.dev
Agentic coding era: Step back and rethink methodologiesNatural language is now the primary programming interface. With the increased use of AI coding agents, there is a strong need for disciplined methodologies. Accountability is non-negotiable. So are coSep 10 · 3 min readtechaarvam.hashnode.dev
Attention is simpler than you thinkPart of the TechAarvam workshop support files — Build Your Own Model. A hand-crafted superhero transformer This notebook presents a hand-crafted example. The goal is to understand the intuition behind3TSep 8 · 12 min readtechaarvam.hashnode.dev
