TALVINDER
  • Work
  • Writing
  • Bets
  • About
    • About Me
    • Library
    • Contact
  • Subscribe
Companies, products, experiments and writing by Talvinder Singh.
Author

B. Talvinder

Builder · Bengaluru · Since 2007

I make things and put them out into the world.

Some become companies. Some stay small. Each one teaches me what to make next.

See what I’m building ↓

Now

What I’m building

Companies I run and products I’m putting into people’s hands.

Companies

01Zopdev

Cloud software for engineering teams.

Company · Current↗
02Ostronaut

Turns conversations on the floor into evidence a business can use.

Company · Building↗
03Pragmatic Leaders

Helps people build the judgment to lead products well.

Company · YC W21 · Coming soon↗

Products

04Pomelo

A recorder that helps you remember what mattered.

Private beta · iPhoneSoon
05Diagrams for Agents

Helps AI agents explain things visually.

Agent tool · Live↗
06P.I.G.E.O.N.

Sends you somewhere worth seeing on the open web.

Internet discovery · Live↗
07PlateLicker

Find the meal worth choosing and get a useful site for every restaurant it recommends.

Food discovery · Live↗
08Elsewhere

Lets work move beyond the machine in front of you.

Open source · Alpha↗
09SourceFoundry

Keeps sources fresh for agents and applications.

Open source · Live↗
10Hukam

A private daily Gurbani reading and reflection practice.

Mac + iPhone · Early release↗
11Side B

A place for people who listen to music with intent.

Music · Shaping↗
There is more, including smaller projects and experiments.See the complete work index ↗

Before

Things I built before

Education · First companyDigital PACEPast Travel · 500 StartupsTushky2011—2015 Travel · Product chapterOYOPast

Writing

Evidence from building AI systems

New collection · Applied AI systemsAI Systems Need an Evidence Layer

How to separate what a system observed, inferred, proposed and was allowed to do.

Read ↗
Release disciplineA Green Build Is Not a Working AI Product↗ Controlled evaluationWhy I Refused to Ship a Better Average↗ Product architectureWhat Changes When Judgment Moves Into the System↗
Frameworks, build logs and field notes.Browse all writing ↗
Title Date
Building the Same Product With AI Changes the Product Contract Sep 5, 2026
A Green Build Is Not a Working AI Product Sep 5, 2026
AI Systems Need an Evidence Layer Sep 5, 2026
Human Approval Is a Product, Not a Button Sep 5, 2026
We Improved 74 of 86 Transcripts and Still Refused to Ship Sep 5, 2026
Agents Need a Source Supply Chain, Not Just Search Sep 5, 2026
Voice AI Is a Chain of Evidence Sep 5, 2026
LangGraph vs CrewAI vs AutoGen: Why Multi-Agent AI Fails on Coordination, Not Capability Jul 9, 2026
RAGs Are Only as Strong as Their Validation Layers Jul 2, 2026
Customization-Scale Paradox Explains Why Vibe Coding Won’t Scale Jul 2, 2026
The AI Bill Will Kill You Before Your Product Does Jul 1, 2026
Why LLM Reliability Must Be Measured Differently in India Jul 1, 2026
The Model Is the Easy Part: What Building Production AI at Every Layer Actually Takes Jul 1, 2026
How We Cut LLM Costs 80% May 11, 2026
How to Monitor AI Agents in Production May 3, 2026
Agentic AI Is Killing Per-Seat SaaS May 3, 2026
The $33B Vertical SaaS Opportunity India Keeps Missing May 3, 2026
Agentic AI Identity Is the Next Frontier in Trust and Compliance Apr 28, 2026
Orchestration Specs Like Symphony Are the Missing Layer for Multi-Agent Engineering Apr 28, 2026
Systematic Large Model Debugging Is the Missing Product Discipline Apr 28, 2026
AI-Assisted Peer Review Is a Feedback Loop Problem Apr 17, 2026
AI Mode in Chrome Is Not Assistantware Apr 17, 2026
Client-Side LLM Optimization Is Misunderstood Apr 17, 2026
Training AI to Serve Rare Disease Patients Is a Structural Problem, Not a Data Problem Apr 17, 2026
The Build-Seed Inversion: Why Marketplaces Die Building Platforms Before Proving Transactions Apr 3, 2026
The Context Window Pricing Collapse Apr 3, 2026
The Martingale Curse: Why Multi-Agent Debates Converge to Mediocrity Apr 3, 2026
The Vibe Coding Hangover: Why AI-Written Code Costs 4x to Maintain by Year Two Apr 3, 2026
The Recourse Trap: Why Competition Makes Credit Scoring More Exclusive, Not Less Apr 3, 2026
Why 86% of AI Agent Pilots Fail Before Reaching Production Apr 3, 2026
I Built Ed-Tech Before Ed-Tech Existed in India Apr 2, 2026
I Built an Experiences Marketplace Five Years Before Airbnb Experiences Apr 2, 2026
The OYO Pivot: When Marketplaces Should Own the Supply Apr 2, 2026
Point-of-Need Learning: Why Application Beats Credentials Apr 2, 2026
The IRL Growth Bet: Why I Started an Offline-First Music Community in the Age of Algorithms Apr 2, 2026
Framework Lag: Why the Winners of 2010-2015 Could Explain Network Effects Before VCs Had Words For It Apr 2, 2026
Capability-Priced Micro-Markets: The Missing Layer Between Agents and HTTP Mar 22, 2026
The Self-Improvement Ceiling: Why AI Can’t Bootstrap Its Way to Capability Mar 22, 2026
Device-Level Blocking Won’t Stop Digital Arrest Scams — The UI Is the Real Vulnerability Mar 22, 2026
Model Routing Is the New Unit Economics Mar 22, 2026
The Small Model Arbitrage: Why India Should Be Building Vertical LLMs, Not Chasing Frontier Mar 21, 2026
We Were Running AI Agents Before ‘Agentic’ Became a Buzzword Mar 20, 2026
The Pass@k Trap: Why Running Your AI Agent 3 Times Makes Answers Worse Mar 19, 2026
AI Is Making Your Team Slower — The Math Your CEO Won’t Show You Mar 18, 2026
Chain-of-Thought Has an Efficiency Tax Mar 18, 2026
The AI Deflation Trap: Why Cheaper Models Make Your Startup Less Defensible Mar 17, 2026
Context Governance at Scale Mar 17, 2026
The OS-Paged Context Engine Mar 17, 2026
The Recursion Threshold Mar 17, 2026
Your AI Agent Is Stateless. That’s Why It Breaks in Production. Mar 16, 2026
Agentic Engineering Is Not Prompt Engineering Mar 16, 2026
Comp Negotiation as an Information Entropy Problem Mar 16, 2026
AI-Generated Code Is Flooding GitHub — Here’s What It Costs You Mar 16, 2026
What Zari-Zardozi Teaches Us About Agent Coordination Mar 16, 2026
Zimbo Meetings and the Ghost Work Tax Mar 16, 2026
Why LLM-as-Judge Fails in India: The $0.03 Evaluation That Costs You Customers Mar 5, 2026
India’s Federated Learning Bet Is a $100M Category Error Mar 5, 2026
AI Infrastructure Will Be Worth More Than AI Applications Mar 4, 2026
If AI can write your PRD, what’s left of your job? Mar 4, 2026
Distress Detection Is a Switching Cost, Not a Safety Feature Mar 4, 2026
Why Indian SaaS Companies Will Win the Agent Reliability War Mar 4, 2026
Software Is Dead. Welcome to the Age of Agentware. Mar 1, 2025
Most AI Companies Will Die the Way Most Cambrian Species Did Feb 20, 2025
Why We Built an Agentic Rightsizing System Instead of Using Existing FinOps Tools Feb 10, 2025
E-commerce 4.0: The Next Evolution of Digital Commerce Nov 13, 2015
Who Killed the Futurist in You? May 31, 2013
Indian Startups Keep Dying at the Social Permission Wall Mar 16, 2013
Entrepreneurship Is Integration, Not Differentiation Feb 25, 2013
Entropy and Entrepreneurship: Borrowings from Science Feb 12, 2013
No matching items

    Newsletter

    New writing by email.

    Frameworks, build logs and field notes. Sent when there is something new.

    © 2026 B. Talvinder

    Work · Writing · Bets · About · Contact

    GitHub · LinkedIn · X


    B. Talvinder is a serial founder (ex-OYO, YC, 500 Startups) who has built four companies across four different technology eras in eighteen years. He writes about AI, infrastructure, and the India tech ecosystem at talvinder.com.

    See a gap in this argument? Say so below.

    Enjoyed this?

    Get frameworks, build logs, and field notes in your inbox.

    No spam. Unsubscribe anytime.