README.md

updated 2 Nov 2026

14 years in startups across growth, product and marketing (smallcase, TeamIndus, Sportskeeda, two of my own). Since January 2026 I write and ship production software myself, by running a team of AI coding agents under a written operating system.

What I build: websites marketing teams can run themselves, brand systems as tools, interactives that explain deep tech, data products from sources with no API, my own products, and small services that replace SaaS glue.

Currently: production engineering on PierSight's people platform; turning scaffold-website into the studio's default starting point; the meeting-intelligence dossier layer.

Contribution activity

Jan–Oct 2026

Commits per month

20
Jan
51
Feb
30
Mar
54
Apr
47
May
121
Jun
22
Jul
56
Aug
157
Sep
68
Oct

626 authored commits in local project repos. Client-repo work (~460 PierSight commits, Aug–Oct) is tracked separately. GitHub contributions: ~1/yr in 2022–24 → 312 in 2026.

Most-committed repos

  • JSX-Renderer (Clariti)200
  • lightcurve-website97
  • polytopia-tracker-web33
  • space-sentinel-website31
  • orbitt-space-website29
  • mandara-capital-website25
  • koyo24
  • orbitt-space-website-live20

Projects

● live = public, returned 200 on 2 Nov

    Lighthouse performance

    run 2 Nov 2026 · score / 100
    SiteMobileDesktopLCP (mobile)
    masonry.studio981002.1 s
    blog.nabhdrishti.in971002.5 s
    Design Forest (review)93982.9 s
    game.piersight.space93992.5 s
    spacesentinel.ai92993.3 s
    Mandara Capital (review)91913.1 s
    lightcurve.ai82984.7 s
    orbitt.space70955.8 s

    Accessibility and best practices score 86–100 on every site, with zero layout shift. Orbitt's mobile score is the Lovable-era SPA waiting on JavaScript; the Next.js rebuild on staging addresses it.

    Case studies

    PierSight: marketing → ops → production engineering

    2023 → present · client

    • Took over the internal admin panel (Aug 2026) and grew it into the HR and people platform.
    • Shipped people records, document vault (server-side masking, 2-min audited reveals), guarded importer with rollback plan, support desk, recognition, business cards, referrals.
    • Agent setup: a worktree per ticket (67), planner + orchestrator + 3 parallel Claude builders, Codex as merge-readiness reviewer. Agents never merge to production.
    • 51 PRs · ~460 commits · 123 Linear issues · 22 ADRs · 5 PRDs · 64 stories.
    NxReactExpressPostgresGoogle CloudClaude CodeCodexLinear

    AI-native comms audit for a fundraise

    Sep 28 – Oct 1, 2026 · deep-tech defence startup

    • Press: 19 items, each verified at the original outlet; aggregators excluded.
    • Video: youtube-pipeline transcribed 10 interviews (2.6 h, Hindi + English, 9 outlets) with speaker labels and claim-focused summaries.
    • Media list of 21 journalists / 210 bylines; 3-month social sample.
    • Output: memo v1–v4, 7 article briefs, a 12-slide HTML pre-read. Every discrepancy checked against the recording before it reached founders.
    Pythonyt-dlpAnthropicCodex in ChromeNotionGoogle Sheets

    Stack & tools

    defaults: Next.js for SEO, Vite + React for tools, Supabase backend
    Languages
    TypeScriptJavaScriptPythonSQLHTML/CSS
    Frontend
    Next.js 16 (App Router)React 19ViteTailwind 4shadcn/uiRadixMotionTipTap
    Backend & data
    SupabasePostgresDrizzleSQLiteExpressPayload CMSZod
    Graphics & motion
    three.jsGSAPCanvasSVGRemotion
    Desktop
    Electronnative macOS (Glaze)
    Hosting & infra
    RailwayCloudflare WorkersCloudflare DNSSupabaseGoogle CloudPorkbunGitHub
    AI models & APIs
    Anthropic APIOpenAI APIGeminiSarvam AIRingg AIgpt-image-2
    Agents & dev tools
    Claude CodeCodexClaude DesignClaude skills & subagentsScheduled routinesgit worktreesVitestPlaywrightWarp
    Integrations
    Notion APIFathomFirecrawlResendLinearyt-dlp

    How I work with agents

    Doctrine

    Every project has AGENTS.md (constitution), PROGRESS.md (shift handoff) and specs before code. 78 projects; a drift audit script; /new-project scaffolds the rest.

    Verification

    "It looks right" is not verification. Browser checks at 375 / 768 / 1280, pixel diffs, tests shown failing without the fix, backtests on real data.

    Parallelism

    ~60 reserved dev-port blocks and a git worktree per task, so agents build at once without colliding.

    Cross-vendor review

    Claude builds, Codex reviews (or the reverse), given the problem without my diagnosis. Nothing unattended near auth, payments, migrations, secrets or infra.

    Gates over prompts

    A prompt is a suggestion; a gate is a gate. Agents never merge to production.

    Instrumented

    Daily summaries → weekly review deck. Claude Code Apr–Aug: ~216 sessions, ~173k lines added.

    Timeline

    1. 2009–2020
      Operator

      Whoopey, Sportskeeda, Cupick, TeamIndus, smallcase (first growth hire → AVP Growth, ~200K → 1M+ users).

    2. 2020–2022
      Fractional growth → Masonry Studio

      Fractional CMO / Head of Growth; studio set up Dec 2021.

    3. 2023–2024
      AI through APIs

      OpenAI API (Apr 2023), Anthropic (Mar 2024). Directed a developer pair on AI tools.

    4. 2025
      App builders

      Supabase, Replit, Lovable. Fractional marketing/ops at PierSight from Oct.

    5. Jan–Mar 2026
      The inflection

      Claude Code, Warp, Pieces. First commit of my own on 5 Jan.

    6. Apr–Oct 2026
      Agent operating system

      Working-with-agents playbook, cross-vendor review, multi-agent crews, production engineering for PierSight.