someshsamanta

6 working demos · all open source · all synthetic data

Marketing systems you can operate, break, and read the source of — before we ever talk.

I build the machinery behind B2B demand generation — capture, attribution, list hygiene, campaign safety, and the reporting that says what actually worked — and then I run it. The six systems below are working reconstructions of production work: each runs in your browser on synthetic data, prints its reasoning, and ships with a test suite. Not screenshots, not case-study prose.

leadflow

ATTRIBUTION · IDENTITY · ROUTING live demosource

One prospect arriving through four channels becomes one CRM record, with every touch attributed to a canonical source by fixed precedence — a paid click ID always beats a stale UTM — and dedup rules that never merge two strangers on a name match.

identity merged into existing contact — arjunmehta@gmail.com ≡ arjun.mehta+vendors@gmail.com (canonicalized); role band filled; nothing overwritten

dailypulse

REPORTING · SLA · HONEST NUMBERS live demosource

Six dashboards compiled into the one email a founder reads: funnel leaks against baseline, leads gone silent past the follow-up promise, ranked actions labeled as hypotheses — and a flat refusal to print a day-over-day delta against stale data.

⚠ NO DAY-OVER-DAY NUMBERS TODAY — yesterday's snapshot is 96h old (max 36h). A quietly wrong number gets believed; fix the snapshot job.

funneldoctor

ANALYTICS FORENSICS live demosource

Before you believe a growth chart, find out how much of it is your own team clicking around. Separates internal pollution from real users, and a real conversion collapse from a tracking event that silently died — with the evidence for every finding.

CRITICAL "export_completed" stopped firing 2026-08-18 — tracking break, not user behavior. Treat export metrics after that date as unmeasured, not zero.

listforge

DELIVERABILITY · LIST HYGIENE live demosource

Segment before you load: live DNS MX checks and catch-all detection split a list into a clean cohort that sends now and a held cohort that gets a human decision — because no verifier at any price can tell a real inbox from a typo on a catch-all domain. The MX panel in the demo runs real DNS, not a script.

Hold — MX (mx.improvmx.com) belongs to a provider known to accept mail for any address by default; per-mailbox verification is meaningless here

safesend

CAMPAIGN SAFETY · COMPLIANCE live demosource

A WhatsApp broadcast engine where the refusals are the product: opt-outs suppressed permanently and checked before capacity, cross-run dedup, hard daily cap, dry-run by default, and a kill switch that halts mid-flight leaving state a resume can trust.

Kill switch pulled after 3 messages. 36 queued contacts were never reached. Resume sends each remaining contact exactly once — no double-sends.

heromatch

CONVERSION · MESSAGE MATCH live demosource

One landing URL whose hero rewrites itself to match the paid click that brought the visitor — search intent, social interruption, or B2B procurement — while organic traffic and search engines always see the one canonical page.

winner click_id:gclid → variant "paid_google" outranked utm_source=newsletter (forwarded-link trap)
THE PATTERN

Every one of these is a refusal layer

A report that refuses stale comparisons. A sender that refuses suppressed contacts. A verifier that refuses to guess on catch-all domains. An attribution engine that refuses to merge strangers. A lander that refuses to show organic visitors a paid promise. The hard part of marketing operations isn't making systems act — it's making them refuse correctly, and print their reasoning so a human can audit the refusal. That judgment is what these demos are actually demonstrating; the code is just how you can check it.

WHO BUILT THIS

Somesh Samanta

Four years in B2B marketing. I started as an SDR — 130+ demos booked cold — then moved into building everything around the sale: positioning, outbound, paid, events, and the CRM and automation underneath. Currently running GTM at a spatial-intelligence platform; before that I built a construction-SaaS demand-generation motion largely solo, and it kept running after I left — which is the standard I hold this work to. Published in NBM&CW (June 2026) alongside the managing directors of JCB India, Sany India, Terex India and CASE Construction.

Stack I ship in: TypeScript, Next.js, Node, Vitest, Vercel · HubSpot, Apollo, Smartlead, n8n · GA4, Google Search Console, Mixpanel.

WHAT HAPPENS NEXT

Hiring for growth, demand gen, or marketing ops? The demos are the portfolio review — email me and we can walk through any of them, including the parts that went wrong in production.

Want one of these running on your stack? Each system exists as a fixed-scope implementation. Tell me which one and what you run, and you'll get scope, timeline and price — not a discovery-call funnel.

someshsamanta6@gmail.com · LinkedIn · GitHub