01 / A daily intelligence agent, not a chatbot

What's changing in Indian youth culture, and what should Lotto, one8 and SportsYard do about it today?

One shareable readout that answers that question every morning. It scans the open web and the marketplaces where the category actually moves, scores what matters, translates it into leadership language, and pushes a summary to WhatsApp. This page is the build plan: what is easy, what is high impact, what we ship first, and how fast.

See the signal map
~2wk
First real signal in leadership's hands
~6wk
Hands-off daily WhatsApp push
~12wk
Full dashboard command center
0custom
Bespoke scrapers needed for v1
02 / The shape of it

Two briefs, one engine

The Daily Intelligence Agent (global brand moves) and the Culture Radar (India youth signal to commerce) are the same pipeline with two output lenses. Build the pipeline once.

01
Scan
Approved public sources, feeds and marketplaces
02
Cluster
Group by theme, kill noise and duplicates
03
Score
Momentum, India fit, sportswear fit, commercial, urgency
04
Summarise
Into leadership language, one line each
05
Recommend
Do now / watch / ignore, per brand
06
Push
WhatsApp summary plus dashboard link
03 / Complexity

The source complexity ladder

Not all sources cost the same to ingest. Here is the honest difficulty of each, from plug-and-play to the ones we phase in later. Tap a tier.

Free signal, the lightest engineering lift

Structured or near-structured sources that need pulling, not fighting. This is where the system earns its keep on day one.

  • Industry RSS and media feeds. Business of Fashion, Vogue Business, Highsnobiety, Hypebeast, WWD, Sneaker News, Complex, Footwear News, The Drum, Retail Dive, plus LBB and Campaign India for the India read. Free, dated, deduplicated by design.
  • Google Trends. Demand and interest curves via pytrends. Free, and the cleanest read on what India is actually searching for.
  • Brand newsrooms. Nike, adidas, Puma, Decathlon, New Balance launch and press pages. One Firecrawl scrape each.
  • Flipkart. The easiest marketplace by a distance. Ready-made actor, HTTP based, no browser, hundreds of products per query, cheap. Public catalog.
Effort to first data
RSS + Trendsvery low
Brand sites (Firecrawl)low
Flipkart (Apify actor)low
VerdictStart here. All of Tier 1 can be live within about two weeks and already produces a credible daily radar.

Worth the extra days, this is the relevant signal

Slightly harder, but this is where the apparel and sportswear signal Agilitas actually cares about lives. Doable with maintained tooling and residential proxies.

  • Myntra, Ajio, Nykaa Fashion. Fashion-first, so the most relevant marketplaces for Lotto and one8 categories. Maintained Apify actors exist, but they need residential proxies to get past anti-bot, so slightly slower and costlier than Flipkart.
  • YouTube. Official Data API for creator coverage, drops and reviews. Quota-limited but fully legitimate.
  • Reddit. Official API across sneaker, Indian fashion and fitness communities. Real community signal, low cost.
Effort to first data
Myntra (Apify + proxies)medium
Ajio / Nykaamedium
YouTube / Reddit APImedium
VerdictBring Myntra in alongside Flipkart in week one. Highest relevance for the brands. Layer YouTube and Reddit in Phase 2.

High value, built once the earlier phases prove out

Genuinely useful and genuinely harder, so not first. Brittle, against terms of service, and a maintenance cost, which is why we build these in a later phase once the core is proven.

  • Instagram and TikTok scraping. Heavily anti-scraping and restricted. For v1, use creator and news coverage as the proxy. The documents already flag this.
  • X / Twitter. API is expensive and limited for this use case.
  • Build it once it is proven. When the core radar is delivering and social listening earns its place, we build that layer in, benchmarked against tools like Brandwatch or Talkwalker.
Effort and fragility
Instagram / TikTok scrapehigh + risky
X / Twitter APIhigh cost
Social listening (build)later phase
VerdictBuild in a later phase. Coverage and marketplace signal carry v1, and we extend into these once that proves out.
04 / Impact vs effort

The signal map

Every candidate source plotted by how much it moves the needle against how hard it is to wire up. The top-left is the fruit worth picking first.

PICK FIRST · high impact, low effort build later LOW EFFORT HIGH EFFORT LOW IMPACT HIGH IMPACT RSS feeds Google Trends Brand sites Flipkart Myntra Ajio · Nykaa YouTube API Reddit API Instagram · TikTok Social listening X / Twitter
Flipkart + Myntra
The marketplace beachhead. Flipkart is easiest, Myntra is most relevant. We are already pulling both and will build out the rest fast.
Pick first
RSS, Google Trends, brand sites. Free, fast, high signal. Live in hours.
Phase 2
Ajio, Nykaa, YouTube, Reddit. Good signal, modest effort, add once the core runs.
Built later, once proven
Instagram, TikTok, X. High value, high effort. We build these going forward, once the earlier phases prove the direction.
The low-hanging fruit, highest impact

Ship the media and Google Trends radar on day one, then add Flipkart for breadth and Myntra for the fashion signal in week one, on scrapers we build in-house. That single combination already answers the daily question for Lotto, one8 and SportsYard, and can be built quickly and shipped fast. Instagram and TikTok come in a later phase, once this proves out.

05 / Sequence

What we ship, in order

Each step is a usable thing on its own. We never build for three weeks and then reveal. Tap through the order.

The daily media radar

Pull the fashion, sneaker and retail feeds plus Google Trends into one place, deduplicate, and you already have a real Daily Brand Radar.

  • What it gives you: top stories on launches, collabs, store concepts and activations across the world, every morning.
  • How: RSS via feedparser plus pytrends, into an Airtable or Sheet, orchestrated in n8n.
Demo-able output
Week 1A dated list of the day's most relevant brand and culture moves, ready to read.

Live commerce signal

Add Flipkart and Myntra so the radar sees not just what brands say, but what is actually selling and trending in the catalog.

  • What it gives you: price points, drops, discount depth and category mix shifts in the exact categories Agilitas plays in.
  • How: ready-made Apify actors for Flipkart (HTTP, cheap) and Myntra (residential proxies), scheduled daily.
Demo-able output
Week 2Marketplace movement feeding the same signal sheet, no custom code.

Scoring and the leadership summary

An LLM clusters the raw signal, removes duplicates, scores each item and rewrites it the way a CEO reads.

  • What it gives you: the top ten, each scored on momentum, India relevance, sportswear fit, commercial potential and urgency, ending in do now / watch / ignore.
  • How: Claude or OpenAI with a fixed scoring and summary prompt. This turns noise into a decision.
Demo-able output
Week 2 to 3The sample signal card at the top of this page, generated automatically.

The WhatsApp push

The summary lands in leadership's WhatsApp every morning, with a link to the full view.

  • What it gives you: the radar where the team already is, at 8am, no app to open.
  • How: WhatsApp Cloud API direct, or Gupshup / WATI / Interakt for easier template handling. Manual for the first days, then automated.
Cost reality
Trivial at this scaleA handful of recipients, one message a day. India rates are among the world's lowest. This is cents, not a budget line.

The dashboard command center

The shareable product: Today, Trends, Competitors, Creators, Brand Actions and Archive, with the heat map and per-brand views.

  • What it gives you: a searchable history, the trend heat map, the brand opportunity view, the creator radar and a weekly auto-generated deck.
  • How: Next.js plus Supabase for the durable build, or Retool / Framer for a faster internal version.
Demo-able output
Week 10–12The full Agilitas Culture Radar, on a link the whole leadership team can open.
06 / Execution

The build, in three phases

The quick wins roll up into three phases, each with a clear finish line and a thing you can put in front of the room.

Signal in the room

A lightly manual but completely real daily radar, fast. The goal is leadership reading actual output within two weeks, not a slide about output.

  • Build: RSS + Google Trends + Flipkart and Myntra via Apify, into Airtable, summarised by the LLM, pasted to WhatsApp for the first few days.
  • Stack: n8n, Apify actors, Claude or OpenAI, Airtable.
Finish line
By week 4A daily Culture Radar message going out, sources and scoring rubric locked, the signal sheet live.

Runs itself

Take the human out of the daily loop. Automate end to end, formalise scoring, and split the view per brand.

  • Build: scheduled pipeline, approved WhatsApp template, full de-duplication and clustering, per-brand views for Lotto, one8 and SportsYard.
  • Stack: WhatsApp Cloud API or a BSP, scheduled n8n, Supabase for the archive.
Finish line
~Week 6A hands-off 8am WhatsApp summary plus a private link, no daily intervention.

The product

The shareable command center the documents describe, with the full section set and a weekly deck generator.

  • Build: Today / Trends / Competitors / Creators / Brand Actions / Archive, heat map, brand opportunity view, creator radar, category filters, search, weekly auto-deck.
  • Stack: Next.js, Supabase, charts, scheduled Firecrawl and Apify. Optional social listening add-on.
Finish line
Week 10+A living dashboard the leadership team and brand owners use to make calls before competitors see the trend.
07 / Build vs buy

The stack, and what makes it fast

Almost none of this is built from scratch. Mature open source and ready-made actors carry most of the weight. Where a license or anti-bot tradeoff matters, it is called out.

Marketplace data
Apify store actors

Maintained Flipkart, Myntra, Ajio and Nykaa scrapers. The single biggest time saver. Data flowing in days, not weeks.

Ready-madeRent + usage
Use what exists, build the rest fast.
Web + brand sites
Firecrawl

Search, scrape and crawl into clean LLM-ready markdown, with anti-bot handled on the cloud tier. Native n8n and MCP support.

Open source~140k★AGPL self-host
Cloud for v1. Self-host triggers AGPL obligations.
Custom scraping
Crawl4AI / Crawlee

For the few sources that need bespoke handling or cheaper scale. Crawl4AI is permissively licensed, Crawlee brings fingerprint and proxy rotation.

Apache-2.0~60k★ / ~22k★
Build later, only if needed.
Trends + feeds
pytrends + RSS

Google Trends interest data and standard RSS feeds. Free, stable, and the highest signal per hour of setup.

Open sourceFree
Build. Trivial and free.
Intelligence layer
Claude / OpenAI

Clustering, de-duplication, scoring and the rewrite into leadership language with do now / watch / ignore. The part that turns a feed into a decision.

APIScales with volume
Buy (API). The core differentiator.
Orchestration
n8n or Make

The daily pipeline as a visual workflow with native nodes for Firecrawl, Apify, the LLM and WhatsApp. n8n is self-hostable.

Fair-codeSelf-host
Build on it. Glue, not code.
Storage + archive
Supabase / Airtable

Airtable to start in days, Supabase (Postgres) as the durable signal archive and the moat over time.

Free tierPostgres
Airtable now, Supabase soon.
Dashboard
Next.js / Retool

Retool or Framer for a fast internal view, Next.js plus Supabase for the durable, shareable command center.

Open sourceVercel
Retool fast, Next.js durable.
Delivery
WhatsApp Cloud API

Meta Cloud API direct (free to host, per-message billing), or Gupshup / WATI / Interakt for easier templates and approvals.

Per-messageIndia rates lowest
Direct, or a BSP for ease.

What it costs

  • WhatsApp: negligible. Internal leadership, one push a day, India among the cheapest markets. Service-window replies are free.
  • Apify actors: modest rental plus usage, the main marketplace data cost.
  • LLM + Firecrawl: scale with volume, controllable by how much you scan.
  • Hosting: free to low tiers on Supabase and Vercel to start.

Where the risk is, and the answer

  • Social scraping: built in a later phase, once the earlier phases prove the direction. Coverage and marketplace data carry v1.
  • Scraper maintenance: outsourced to maintained Apify actors, not in-house code.
  • Noise over signal: handled by scoring and the do now / watch / ignore discipline.
  • License exposure: Firecrawl cloud or Apache-licensed Crawl4AI, never AGPL self-host in the product.
08 / Phased

One quarter, first signal by week two

Built so there is something real in leadership's hands early, with the full product landing inside a quarter.

Week 1
Sources, fields and scoring locked
Apify actors configured, n8n skeleton, feeds wired.
Week 2
First radar in WhatsApp
First AI summary, first message to leadership. Visible output.
Week 4
Phase 1 complete
Daily summary running, Flipkart and Myntra signal flowing.
Week 6
Fully automated
Hands-off 8am push, scoring formalised, per-brand views.
Week 10–12
Dashboard live
Full command center plus weekly auto-generated deck.
~2 weeks to first real output ~6 weeks to hands-off daily ~10–12 weeks to full dashboard 0 custom scrapers for v1
09 / For the brainstorm

Six decisions to make in the room

This is a draft to argue with, not a finished spec. These are the calls that set the build going the moment they are made.

Which brand leads v1?
Lotto, one8 or SportsYard. Lean: one8, given the relaunch, the Kohli story and the global push, with Lotto close behind.
Which eight to ten sources do we lock for day one?
From Tier 1 plus Flipkart and Myntra. Pick the exact feeds in the room.
Dashboard: fast or durable?
Retool or Framer to move quickly, or Next.js for the long-term product. Can start fast and migrate.
Who receives the WhatsApp, and how?
A leadership group or named individuals, and which number sends it.
Where does the data live, and who owns it?
The signal archive compounds into a real asset over time. Decide hosting and ownership now.
Self-host or cloud for scraping?
The anti-bot, cost and license tradeoff. Recommend cloud and managed actors for speed in v1.
01 / 09 · Overview