Video summary

10 Projects I've Built and 1 Thing I Learned From Each

Main summary

Key takeaways

Business

Business-focused summary (strategy, ops, leadership, marketing, product)

Core theme / “playbook”

  • Entrepreneurship ≠ engineering: success depends more on distribution, sales, and go-to-market timing than on shipping features.
  • Run experiments, don’t get attached: treat “startup” work as a series of validated experiments; some will fail, but failures reveal what will work.
  • Avoid/defeat competition by timing: aim for a 1–2 year window with low/no direct competition to build pricing power and foothold.
  • Distribution-led product: build/market around how users discover and share; don’t assume “cool tech” creates demand automatically.

Project-by-project: key execution lessons + outcomes

10) Jobspire (job board) — “too technical, not enough sales + flawed revenue model”

What they did

  • Built a job board for recruitment; raised ₹1.7 crore while in college.
  • Entered an incubator and shifted from Manipal to Gurgaon.
  • Built payments infrastructure in an era with no simple third-party gateways (e.g., no Stripe/Razorpay).
  • Value proposition: charge only when hired (performance-based).

Problems

  • No demand early: first companies weren’t converting.
  • Misuse of incentives: companies could “get hired” or bypass the intent (e.g., restarting directly off-platform), undermining the take-rate model.
  • Team + culture bias: overly technical; excitement for features over sales/process.
  • “VC learning by doing”: had to learn without mainstream entrepreneurship resources.

Learning / would do differently

  • Be more sales-focused earlier:
    • move from “a few recruiters” → “thousands of recruiters/employers”
  • Shift engineering energy toward growth loops and GTM execution, not just platform build-out.

Actionable recommendation

  • Validate your business model incentives early (don’t just test functionality).
  • Build a sales motion before the product is “feature-complete”.

9) Avalon Scenes (consumer chat app → B2B pivot → sell)

What they did

  • Built a consumer messaging/live community app during COVID; reached hundreds of thousands of users.
  • Strong engagement:
    • rooms of 200–500 people
    • room with ~1,000 people
    • user “usage” and community density were strong
  • Pivoted to B2B to survive:
    • ~3–4 months to make it B2B-ready
    • built an SDK (their own) enabling “your own Discord on your platform”
    • added extras like forums, but adoption didn’t follow

Why it struggled

  • B2B customers were not the intended winners:
    • early customers: creators wanting communities
    • later customers: “boring companies/banks” who couldn’t replicate creator-driven distribution
  • Platform substitution risk:
    • when WhatsApp community emerged, creator audiences moved there
    • limited differentiation because scale/attention already existed on WhatsApp

Key product/engineering mistakes

  • Over-engineering core tech:
    • built their own WebRTC live calling solution over 6–8 months
    • then multiple ready-made SDKs appeared soon after (e.g., 20MS mentioned)
    • wasted engineering cost (estimated “a few crores”)

Learning / would do differently

  • Prioritize build-vs-buy:
    • if commoditization is likely, don’t build deep infrastructure unless it’s your strategic moat
  • Don’t rely on customer requests alone:
    • customers praised features but defected to WhatsApp due to user base and onboarding friction
  • Understand:
    • stated preferences vs revealed preferences
    • “faster horses” problem (Henry Ford analogy)

Outcome

  • Ran ~8 years for Avalon; Scenes/B2B version ran ~4 years
  • Generated “a few crores revenue per year” at one point
  • Sold to Unacademy

5) “God in a Box” (WhatsApp + ChatGPT API) — “timing + creator-driven traction”

What they did

  • Shipped quickly: launched 2–3 days after the GPT API became available.
  • Go-to-market channel: WhatsApp wrapper (built payments + simple auth + GPT wrapper).
  • Convenience first: paid WhatsApp fees via intermediaries; focused on onboarding ease.

Key metrics / growth

  • Cost: GPT API was “crazy” initially, then 10x cheaper within ~2 months
  • Users: 0 → 1.4 million
  • Revenue: became profitable (before monetization collapse)
  • Geography: viral in Spain, Cuba, South America (influencer-led spread)

Marketing motion

  • Marketing-first, product-second:
    • Tweeted: “launching—retweet for early access”
    • got ~4,000 retweets

Inflection point logic

  • Tech hype created a demand window (they rode the inflection point).

Negative

  • Revenue crashed after Meta/ChatGPT AI entered WhatsApp, removing differentiation and causing users to churn.

Learning / would do differently

  • Build in new areas quickly during hype windows.
  • Traction beats feature perfection.
  • Add value via convenience, even if the tech layer is “simple”.

5) AI Avatars — “new business model beats commoditized software”

What they did

  • Identified a business shift:
    • credits-based software model was weak/commoditizing
    • use AI avatars to solve a real creator pain: exhaustion / consistency
  • Product claim: became the largest AI avatar globally (recognized around WEF Davos early 2025).
  • Output scale: ~100 million views per month
  • Business model:
    • revenue mainly from influencer marketing, not direct software charges
    • spend went to AI tooling; monetization came from campaigns/ads/influencer distribution

Competitive moat

  • Innovation wasn’t only in the tech layer:
    • business model innovation layer
  • Shifted to custom model (fine-tuned LoRA / “VAN 2.1”) after others commoditized.

Learning

  • When the product commoditizes, win by:
    • packaging into services
    • distribution/media strategy
    • brand positioning + channel count

4) Autocode Pro — “run experiments, but know when not to compete with big tech”

What they did

  • Built as an experiment (no major direct revenue).
  • Competed in an “AI coding assistant” direction, using a code-generation repo/extension-based approach.

Reason it didn’t become the main business

  • Expected competition escalation quickly:
    • big players (OpenAI/Microsoft/Google/Claude) move fast
  • Funding reality:
    • required “minimum $100M” style resources (their view)
  • Decision: keep it an experiment rather than a high-risk platform war.

Learning

  • Startup = experiment; not every promising tech direction is winnable without scale.
  • “Knowing when not to fight” is a core leadership skill.

3) Video Vault — enterprise on-prem moat (hardware + permissions + services)

What it is

  • “GitHub for video” (versioning/comparison and collaboration for large media files).
  • Problem addressed:
    • editing pipelines with huge assets (e.g., 200–300GB raw data per video/company workflow)
    • cloud upload/download costs and workflow friction

Strategy

  • Choose on-prem to create a moat:
    • built on TrueNAS + Samba
    • heavy permissioning, backups, LAN setup complexity
    • moat is mostly hardware + integration/systems work they don’t directly sell; they enable it

Scale metric

  • Processes 2,000+ videos/month (on-prem setup for enterprise)

Positioning

  • Enterprise-directed messaging
  • Sometimes bundled with broader content services

Learning

  • Even as “software becomes only part of the offering,” defensibility can come from:
    • operational complexity
    • infrastructure integration
    • on-site deployment + services layer

2) Unleash the Avatar (game project) — “distribution lens + global narrative + country constraints”

What they did

  • Building a Souls-like / Sekiro-like game:
    • described as having a ~40-person team
    • custom Unreal Engine/modded enemy AI + combat systems (not GenAI-based for AI)
  • Distribution strategy:
    • China-first due to broader PC gaming audience
    • heavy organic reach from trailers and content creator lens
  • Production workflow:
    • photogrammetry scans of real 13th century assets (team sent to Chanderi)

Marketing + traction metrics (China)

  • Trailer outcomes:
    • first trailer: mixed feedback
    • second trailer: hundreds of millions of views same day
  • Status:
    • 6th most played video on IGN China
    • ~10–20k views below GTA 6 trailer on that chart (as stated)
  • No marketing spend:
    • “Until today, we haven’t spent one rupee/dollar on marketing”—organic

Unit economics insight

  • Country constraint: India lacks enough PC gaming audience, raising international sales cost.
  • Example cited:
    • 70%+ of Wukong revenue came from China’s home market, hard for India to replicate

Learning

  • Global fan traction + meme culture can change outcomes (tracked narrative evolution publicly).
  • Talent hiring got easier after public trailers (early skepticism → later interest).

1) EOS (systems + leadership) — “build a platform that multiplies talent”

What EOS is positioned as

  • Not another product; a leadership/execution system to enable many projects by young teams.
  • EOS goal:
    • let young, high-agency talent build the next Jobspire / Unleash the Avatar / Scenes, etc.
  • Emphasis:
    • share lessons internally daily
    • bootstrapped leadership team success

Learning

  • Long-run advantage:
    • accumulated capital + distribution + a culture/system for experimentation
    • leadership pipeline that compounds over time

Frameworks / playbooks explicitly or implicitly used

  • Stated vs revealed preferences
    • Customers request features, but adoption follows where users + convenience already exist (WhatsApp case).
  • Distribution-first product thinking
    • “Marketing first, product second” (God in a Box).
  • Experimentation / “startup = experiment”
    • run public experiments; learn from both success and failure
    • maintain a “graveyard of experiments” to avoid repeating mistakes
  • Moats through implementation complexity
    • Video Vault: on-prem hardware + TrueNAS/Samba + backup/permissioning + services
  • Competition timing / avoid zero pricing power
    • target 1–2 years with low/no direct competition
  • Henry Ford “faster horses”
    • customers optimize what they ask for, not what solves the underlying problem

Concrete KPIs / metrics mentioned (and relevance)

  • Jobspire
    • Raised: ₹1.7 crore
    • Investor outreach: emailed 70 investors (69 no, 1 yes)
  • Avalon Scenes
    • Users: hundreds of thousands
    • Engagement: rooms 200–500, sometimes ~1,000
    • Engineering effort: 6–8 months for live calling infra
    • Pivot effort: 3–4 months to become B2B-ready
    • Revenue: “few crores per year” (exact figure not specified)
  • God in a Box
    • GPT API cost drop: 10x within ~2 months
    • Users: 0 → 1.4 million
    • Engagement: ~4,000 retweets from launch tweet
    • Revenue driver: profitability before monetization collapse; crash after WhatsApp integration by Meta/ChatGPT
  • AI Avatars
    • Scale: ~100M views/month
    • Recognition: “largest AI avatar in the world” (WEF Davos early 2025)
  • Video Vault
    • Throughput: 2,000+ videos/month
    • Data scale: 200–300GB raw data per video workflow
  • Unleash the Avatar (game)
    • Team size: ~40 people
    • Trailer impact: “hundreds of millions of views same day” (second trailer)
    • Ranking: #6 most played on IGN China (out of 20,000 videos)
    • Marketing spend: $0 / rupee 0 on marketing (as stated)
  • EOS
    • No explicit numeric KPIs provided; focus is organizational outcomes (bootstrapped leadership and enabling projects)

Presenters / sources

  • Presenter (unnamed in subtitles): referenced via examples including “Varun May” (used as a discount/sponsor-code marker) and as an audience marketer in examples.
  • Company/source mentioned: Purvi Capital (VC)
  • Brand/examples mentioned: Zomato, Hike, Razorpay, Stripe, 20MS, WhatsApp, Unacademy, Morph (modelcode.ai), Henry Ford quote, Peter Thiel quote, Frame.io, TrueNAS, Samba, Wave2Lip, HeyGen, Synthesia, dev names (e.g., Varun May), Google I/O, Meta Connect, World Economic Forum (WEF), IGN China, GTA 6, Wukong (game), Asmongold, Bilibili, YouTube/AV contexts, EOSCompany.com (EOS)

Original video