Take the power
back.
Contra.Ventures is a product studio of one. Tools that return agency to the person holding the device — not the platform behind it. Built in Bangalore by Param Vaswani, with AI as the leverage and zero permission to wait for.
Bet on your body. Commit with your wallet. Think in reverse. Wire yourself different. Take the power back.
Shipping Keep oracle v2. Drafting first long-form on oracle-as-moat.
Whoop 17% — deep red. Catching up on sleep tonight before the next sprint.
Wendell Berry · Hannah Coulter. Patrick Collison threads.
Contra
means
against.
against extraction · against resignation · against the quiet assumption that someone else will fix it
We were promised the future. They handed us the engagement metric.
Twenty years of platform capitalism gave us a feed that runs on outrage, an attention economy that monetized the inside of our skulls, and a pile of products none of us would have asked for if we'd been asked. The prevailing mood is gratitude.
Contra.Ventures starts from the opposite premise. The internet was supposed to be a liberation technology. It still can be. What it needs is people willing to build with that intent — not as customers, not as creators inside someone else's walls, but as owners.
A studio of one is a contradiction in terms. That's the point. With AI as a leverage multiplier, one person can hold what used to take a team of ten. Tighter loops. No meetings. No political capital. No quarterly review. Just the work, the ship date, and the next idea.
Every tool here traces back to one stake — return the lever to the person holding the device. If a product extracts your attention without giving you compounding agency, it's a platform. If a product hands you a lever you didn't have before, it's a tool. I build tools.
Bet on your body, not the algorithm's read of it. Commit with your wallet, not a screenshot you'll forget. Think in reverse, because forward-only is how the platform thinks. Wire yourself different, because the default wiring was somebody else's project plan.
Take the power back. Not by opting out — that's the easy fantasy. By opting in to the only world we have, and rebuilding it from the inside, one tool at a time.
“If a product extracts your attention without giving you compounding agency, it's a platform. If a product hands you a lever you didn't have before, it's a tool. I build tools.”
I deleted the
standard plan.
Bangalore, early 2026. Solo, no cofounder, no warm intro, no incubator.
Started Contra.Ventures in early 2026, in Bangalore, solo. No co-founder, no warm intro, no incubator. The plan was the standard plan: raise pre-seed, hire two engineers, ship in eighteen months, hope you don't run out before product-market fit lands.
I wrote that plan. Then I deleted it.
What changed: AI got good enough that one person plus a Claude Code window could sustain the cadence of a small team. So I started shipping. Keep first. Then BONP, the oracle protocol underneath it. Then Param Hub for the data, Wired Different for the community, a half-dozen smaller experiments. By month three, twelve products live, four MCP servers published, six evals open-sourced. All on GitHub. All settled by data my body produced.
Contra.Ventures is the holding shape for the whole portfolio. A studio means I make things, I ship them, and the throughline is consistent enough to call it a body of work. Not a startup pretending to be a movement. Not a movement pretending to be a startup. A studio.
The thesis I'm betting on: the next decade rewards people who can hold the entire stack — taste, distribution, code, content, biometrics, AI — alone or in tiny teams. The age of the indie maker is back, and this time the leverage is real.
Keep.
Bet on yourself.
Charity-forfeit commitment markets settled by your Whoop or Oura. Bet on yourself. Miss, you fund the cause you despise.
Keep is what happens when you stop trusting your willpower and start trusting infrastructure. Set a goal — eight hours of sleep, 75% recovery, a sub-seven mile. Stake money behind it. Miss, and the money goes to a charity you actively dislike. Your Whoop or Oura settles the market. No self-reports. No loopholes. No I'll-start-Monday. Beeminder's idea, with real markets, real stakes, and a cryptographic oracle underneath. Pre-seed, play-money live, Gibraltar-licensed real money in 6–18 months. The thesis: the most underused asset in your life is the data your body is already generating.
Shipped.
Live and in use. Some pay rent, most feed the loop.
Both, And
wipSocial and intellectual matching for people who hold contradictions.
relationship with Keep TBD
“A studio of one is a contradiction in terms. That's the point.”
Measured.
Benchmarks I couldn't find, so I built them.
CMCHP
researchCross-Model Context Handoff Protocol v0.1.
A spec for passing intent between different LLMs without re-priming. Paper plus benchmark.
paramxclaudedev/cross-model-handoffMCP Tool-Selection Eval
researchHow accurately do models pick the right tool from a menu of 78?
7 models, n=665. Opus 4.7 = Sonnet 4.6 = 97.7%. GPT-5 Mini beat GPT-5.
cross-provider · open-sourcingbiometric-json-eval
research29-case benchmark for settling biometric markets with LLMs.
Opus 4.7 + Gemini quorum produced ~0 dangerous errors at ~$0.05 per settlement.
Voice-Preservation Eval
liveHow many rewrites before a model destroys your voice?
Sequential rewrite chain per model. Opus judges drift each iteration.
~/.claude/skills/voice-preservation-evalClaude-Code Hook Eval
liveActual wall-clock time saved per hook in my settings.
Parses transcripts, counts fires, benchmarks latency. Outputs a keep/tune/drop verdict.
~/.claude/skills/claude-code-hook-evalyt-tldr-eval
research100-video YouTube summarization bake-off. Opus judges.
The machinery.
MCPs, hooks, agents. The scaffolding behind 122 commits a week.
whoop-mcp
liveWhoop API as an MCP server. Public on GitHub. Published to npm.
paramxclaudedev/whoop-mcpspaceship-mcp
live14 tools for the Spaceship domain registrar.
vercel-mcp
live15 tools for the Vercel REST API.
claude-code-hooks-mcp
liveRead-only MCP for inspecting ~/.claude/settings.json hooks.
Claude Code Playbook
live4 MCPs, 11 skills, 15 hooks, 5 evals. A public monorepo.
paramxclaudedev/claude-code-playbookLife OS
liveSeven autonomous scheduled agents running my day.
Morning brief. Keep monitor. Friday lessons. Monthly synthesis. Inbox triage. Cron + launchd.
paramxclaudedev/life-osCompaction Predictor
liveA hook that predicts Claude Code context compaction before it happens.
Hybrid hard-rule plus logistic regression. Active in my settings.
paramxclaudedev/context-compaction-predictorInbox Triage
livelaunchd agent routing my Todoist inbox every 30 minutes.
Biometric Inference Proxy
wipHono proxy. Whoop recovery → four modes → rewrites temperature and thinking.
mcp-red-team
wipAdversarial MCP security scanner. Eight attack categories.
Zero
willpower
on logistics.
Wake up. Whoop tells me what kind of day this is. Recovery green ships code. Yellow writes. Red reads.
A morning brief runs at 7am via launchd — pulls Whoop data, Notion tasks, calendar, GitHub commits, and prints a single page of what to do. By 9am I'm in a Claude Code window. By noon something has shipped or a draft has gone live.
The stack: Claude Code as the primary IDE. Cursor for visual work. Custom MCP servers for everything I touch repeatedly — Whoop, Spaceship, Vercel, Notion, Todoist. Hooks for compaction, formatting, sound cues. Skills for repeating workflows. Memory files so the system remembers me between sessions.
Output, last seven days: 122 commits across 36+ repos. Three to five long-form pieces a month. One new product or eval per week.
Inputs that move the needle: Wendell Berry, Hunter Thompson, Beeminder's archive, Patrick Collison's signal-boost feed, every Anthropic paper that drops. Sleep, lifting, walking, repeat. The point of the system isn't speed for its own sake. It's spending zero willpower on logistics so all the willpower lands on taste decisions, hard writing, and the next bet.
“The age of the indie maker is back. This time the leverage is real.”
Watch the studio
in motion.
Param Hub is my Life OS — the public dashboard where everything I track lives. Whoop recovery, today's tasks, sleep, training, content pipeline, the works. Updated continuously, no curation. If you want the unfiltered view, it's there.
Currently.
What’s in my head right now. Nothing aspirational, all in motion.
- Building
Keep oracle v2. biometric-json eval going public. YC application for May 4.
- Writing
First Substack on oracle-as-moat. Drafts on solo-founder leverage and ADHD-as-spec.
- Reading
Wendell Berry, Hannah Coulter. Latest Patrick Collison threads on progress. Constitutional AI papers.
- Recovering from
Whoop 17%. HRV 29ms. Sleeping more is the only honest answer.
- Listening to
Mom's House for stupid. Conversations with Tyler for serious.
- Stake
First long-form piece live by the end of this week. No more drafts.
Receipts.
Dated, public, irreversible. Skip the pitch — read the diff.
- APR 24shipped
Whoop OAuth complete
Keep oracle v1 fully wired. Real biometric data settling real markets.
- APR 21open-sourced
150 skills installed
Writing, research, scientific, finance, ops — the full Claude Code surface.
- APR 20shipped
param-site live
Single-page consolidation portfolio at param-vaswani.vercel.app. Now reborn as contra.ventures.
- APR 19shipped
Inbox triage agent shipped
launchd, every 30 minutes, Sonnet 4.6 routes Todoist inbox automatically.
- APR 19shipped
14-agent parallel sweep
Auto-mode fan-out: 14 non-overlapping subagents, ~14/14 completion.
- APR 17shipped
Lifebet → Keep pivot
Charity-forfeit commitment markets become the thesis. 90% focus.
- APR 17evaluated
biometric-json eval
29-case benchmark. Opus 4.7 + Gemini quorum hit ~0 dangerous errors at $0.05/settlement.
- APR 15evaluated
MCP Tool-Selection eval v2
7 models, n=665 cross-provider. Opus 4.7 = Sonnet 4.6 = 97.7%. GPT-5 Mini beat GPT-5.
full history on github. live state at likehearted.life.
The week,
on rails.
Modes by default. Whoop overrides them when the body says otherwise.
- Monshipdeepest code work
- Tueshipevals + research
- Wedwritelong-form
- Thushipinvestor / partner calls
- Frisynthweekly lessons + planning
- Satresttraining + reading
- Sundriftno schedule
ship · write · synth · rest · drift — rotated weekly, audited monthly
Ideas I
bet on.
Load-bearing convictions. Every build traces back to one.
- 01
Biometric sovereignty
Your body generates more data than any platform. It should settle your bets, gate your commitments, and stay encrypted. Oracles, not self-reports.
- 02
Solo is a structural advantage
With AI as a force multiplier, one person can run what used to need a team of ten. Fewer meetings, tighter loops, clearer taste.
- 03
Commitment beats motivation
Willpower is weather. Contracts with your wallet, your body, and your friends are climate. Build the climate.
- 04
Tools, not platforms
Platforms extract attention. Tools return agency. I build tools.
- 05
Ship before permission
The internet rewards the people who were already shipping when the moment arrived. Don't wait to be anointed.
- 06
ADHD is a spec, not a bug
Tools designed for neurotypical focus quietly exclude half the people who could use them best. Design for the brain you actually have.
Shelf
of
influence.
Books, essays, podcasts, papers. Curated, not recommended.
- bookWendell Berry
Hannah Coulter
On membership, locality, and the slow life. The antidote to platform-scale thinking.
- bookHunter S. Thompson
Fear and Loathing in Las Vegas
The voice. First-person, full velocity, no apologies.
- essayDaniel Reeves & Bethany Soule
The Beeminder archive
Twelve years of writing about commitment devices. Keep stands on this corpus.
- podTyler Cowen
Conversations with Tyler
The most demanding interviewer in tech-adjacent media. Calibrates taste.
- paperAnthropic
Constitutional AI
The substrate I build on. Worth re-reading every quarter.
- newsletterJerusalem Demsas et al.
The Argument
Style bible for my long-form. 16 essays studied, voice sampled, prose tightened.
- bookPatrick Collison & co.
Stripe Press catalog
The bar for what a tech book can be. Permanent reference shelf.
- podTom Segura & Christina P.
Mom's House
Pure stupidity for the recovery days. ADHD brain needs both gears.
Voices in
the room.
Some are people. Some are corpora. All of them get a chair at the table.
YC + a16z + AI Grant founders
Applying May 4. The bar there sets the bar for Keep — every rejection sharpens the pitch, every yes opens a different door.
Solo-founder Slack circles
Quiet rooms of people doing what I'm doing. The ones who don't post about it.
ADHD-coded operators
Folks who built with the brain they have, not the one they were told to want. We trade tools.
Causes
I care
about.
Return agency to the individual.
The default has been: platforms set the terms, users comply. I'm building the opposite. Every tool here tries to give the user back a lever.
Make the frontier legible.
AI is moving faster than most people's mental model of it. Writing, evals, and open playbooks narrow the gap.
Bridge the tech-human divide.
We don't get to opt out of the networked world. We do get to reclaim how it's built. Contra means against — against resignation, against extraction, against the idea that someone else will fix it.
What I
actually do.
- product strategy
- full-stack (TypeScript · Next.js · Python)
- AI systems (Claude, Gemini, OpenAI, agentic tooling)
- MCP servers + Claude Code hooks
- prediction-market design
- biometric data pipelines
- evaluation & benchmarking
- narrative writing & GTM content
- solo founding & fast shipping
Honest
answers.
The questions I get on every call. Saving you the email.
Why solo? Why not raise and hire?
+Because the leverage shifted. With AI, one person plus a Claude Code window can hold the cadence of a team. Hiring before product-market fit is how good ideas get diluted. I'll hire when the ceiling is the bottleneck, not before.
Why Bangalore?
+Because I live here. Because India is the most under-told tech story of the next decade. Because the cost-of-living arbitrage means I get to build for two years on what'd buy six months in San Francisco. And because the talent pool is closer than the discourse admits.
Why publish everything on GitHub?
+Because moats from secrecy are weaker than moats from execution. The MCPs, the hooks, the evals — they're all open. The thing that matters is the rate of compound on top of them, and that's not transferable.
How is Keep different from Beeminder or StickK?
+Beeminder is the prior art and I cite it constantly. The differences: Keep settles via biometric oracle (no self-reports), uses real markets (not just commitment contracts), and routes forfeits to charities you actively dislike (anti-charities), which research suggests is meaningfully more motivating than neutral ones.
Are you actually shipping a real-money product?
+Play-money is live now. Real money requires a license — the Gibraltar track is the path. Six to eighteen months. Until then, the play-money version compounds users, surfaces UX bugs, and proves the oracle works.
What's the studio's business model?
+Keep has a clear path: take rate on settled markets. Everything else in the studio is either funded by Keep, monetized via Substack/courses (Wired Different), or kept open as research that compounds the studio's surface area. No SaaS-by-default.
Specific
asks.
Vague networking is a tax on everyone. Here’s exactly who I want to hear from.
- Investors / pre-seed checks
Building Keep toward a $1.5–2.5M pre-seed. Biometric oracles, charity-forfeit commitment markets. Real-money license track in Gibraltar.
- Design partners
If you'd actually use Keep — daily — to commit to a hard goal, I want to talk. The first hundred users shape the product.
- Solo-founder peers
Trading playbooks with people running studios of one. Not advisor calls. Real working sessions.
- Long-form editors
The essay pipeline is real and shipping. If you commission contrarian-tech writing, the door is open.
- YC / a16z / AI Grant alumni
Applications going in over the next two weeks. If you've been through the process and have ten minutes for sharper feedback on the pitch, I'll send the deck.
If this sounds like
your world —
Investors, operators, writers, solo builders, anyone betting on their body, their brain, or their time. Write directly. No forms.