Latest ·
What Dussehra knows about building with AI
Ravana knew everything and still lost. Dussehra's ten heads, the tool worship and the burning effigy read well as a yearly review for anyone shipping software with AI.
Kalpak Kale
Dispatches on putting AI to work · evaluations, agent workflows, shipping without a dev team
Kalpak Kale is Operations Manager at Latent, an Alpha OBS LLP venture, based in Maharashtra, India. With over twelve years across media production, IT infrastructure, project management and operations, he focuses on AI implementation, automation and software delivery: enterprise RAG chatbots, agentic coding workflows, AI tooling adoption and production application development. Everything here is written from work that shipped; client and product names are left out, the numbers are real.
See the portfolio Get in touch
Also on LinkedIn, Medium, dev.to and Hashnode. Posts here are the canonical versions. RSS.
Latest dispatch
All dispatches
Thoughts vs thinking: the distinction that quietened my mind
Thoughts vs thinking: what the Bhagavad Gita says about the difference and three practical ways to quiet an overactive mind.
Every message is a seed: designing an intent-first RAG chatbot
Design pattern for trustworthy chatbots: treat each message as a seed, context as soil, retrieval as nourishment, turn values into testable response rules.
The 21-case test that decides whether a RAG chatbot ships
The 21-case test that decides whether a RAG chatbot ships: weighted evaluation, nine QA rounds, and client-run daily accuracy reports.
Giving 19 developers AI access through one proxy, and what actually changed
Rolling AI coding tools out to a team is not just licensing. It is a gateway, spending policy, evaluation habit and working rules for a 60-person firm.
A brief goes in, software comes out. Shipping client products without a dev team assigned.
An operations manager with an agentic coding tool can ship a statement of work to production using a fixed process for payments and security.
Written by hand, one every couple of weeks. Browse the archive.
News update Daily news updates
One item a day on a real AI innovation, researched and drafted automatically from primary sources, always marked as a news update. All news updates.
Moonshot AI Kimi K3.1 Pro nears frontier, open weights due Nov 4
Moonshot AI's Kimi K3.1 Pro scores 65.7 percent on Terminal-Bench 4.0, closes in on Claude Opus 5.5, and promises open weights by November 4.
NVIDIA opens OpenShell, a safe runtime for autonomous agent fleets
NVIDIA's OpenShell gives agents file, package, API and credential access under kernel enforced policy and formal verification.
Recursive launches Flint, an open local coding model for agents
Recursive released Flint, an open-weight 40B MoE coding model that runs on laptops and scores 74.2% on SWE-Bench Verified, built largely by AI.
Google and Meta open-source TorchTPU for native PyTorch on TPUs
Google and Meta have released TorchTPU, a native PyTorch backend for TPUs. It removes the PyTorch/XLA layer and adds eager mode and torch.compile support.
MiniMax M3.1 Flash undercuts DeepSeek on cost and coding tasks
MiniMax rolled out M3.1 Flash, a 428B parameter MoE agentic model with 1M context, open weights due October 23, and coding tasks at half the cost.