The process

The portal is the
proof of work.

Every page in this portal was built from scratch — real data, real infrastructure, real code. This page explains exactly how. No templates. No consultants. No fabricated numbers.

How it was built — step by step
01
Pulled Cresta's brand kit and ICP definition
Started with context.dev's brand API to extract Cresta's actual color system, logo, and typography. Read every public source: the Forrester Wave report, the Gartner Magic Quadrant, case studies from United Airlines and Optimum, and Ping Wu's public LinkedIn posts. Built the ICP definition from scratch: enterprise contact centers (5,000+ employees), running Genesys/Avaya/NICE/Five9, in FSI, Telecom, Airlines, or Healthcare.

No guesswork. ICP = what Cresta has already won.
Research
02
Analyzed 461 Ping Wu posts with a Python scraper
Ran a Playwright-based LinkedIn scraper against Ping Wu's full post history — extracted dates, engagement, post type, and content for all 461 posts. Used a CSV analysis pipeline to identify the 8 core voice fingerprints: contrarian takes, data-first, short paragraphs, no filler words, always ties to a real outcome. Built the voice system in Rock 02 directly from the analysis output.

Also scraped the top 3 competitor founders (Genesys, NICE, Five9) for competitive voice positioning.
Python · Playwright · LinkedIn Scraper
03
Ran 63 contact centers through the Mystery Shopper system
Built a Vapi-powered phone scanner that dials each account's support line, navigates the IVR tree, and returns structured intelligence. The scanner runs via TypeScript activities on a Temporal workflow — durable, resumable, and fully logged. Analyzed 63 calls and extracted the data in Rock 01: 58% no AI at the front door, 71% run 4+ IVR levels, 86% on legacy platforms.

This is not simulated data. The calls happened. The transcripts were analyzed by Claude.
Vapi · TypeScript · Temporal · Claude
04
Sourced 50 accounts from 8 live signal sources
Ran account qualification against TheirStack (174M job postings), SEC EDGAR (10-K filings), Grok (X/Twitter real-time signals), People Data Labs (contact enrichment), The Swarm (warm intro mapping), Listen Notes (podcast discovery), and the Mystery Shopper results. Built a compound scoring algorithm: TheirStack Genesys + SEC attrition risk + new VP CX hire = auto-escalate to critical.

Heat scores are computed, not assigned. Every number traces back to a source.
TheirStack · SEC · Grok · PDL · Swarm
05
Mapped 40 contacts via Vayne.io → FullEnrich pipeline
For each account's target contacts (CCO, VP CX, SVP CS), ran LinkedIn profiles through Vayne.io to extract profile data, then through FullEnrich.com to enrich with email addresses and phone numbers. No human in the loop — the entire contact discovery pipeline runs automatically given the account list.

The only human checkpoint was approving the 50-account target list. Everything after that ran without human involvement.
Vayne.io · FullEnrich · Python
06
Generated sequences in Ping Wu's voice using the voice system
Fed each contact's signal data (stack, Mystery Shopper results, job posting activity, Grok signals) along with the Ping Wu voice system into a sequence generator. Output: 10 three-touch sequences — Day 0 opens with the Mystery Shopper finding as the hook, Day 5 is the ROI math, Day 12 is the social proof exit.

Touch 1 is different for every account because the IVR call data is different for every account.
Claude · Ping Wu voice system · signals.ts
07
Built the portal and deployed it
Designed the portal to match the structure of a system Cresta's GTM team would actually want to run — not a pitch deck. Built with HTML/CSS/JS (Inter font, Cresta brand tokens, Ramp-inspired dark design system), deployed to a Nginx server with SSL. The login gate uses session storage — no backend auth, no overhead.

The portal is the demo. Every feature you see is a feature I'd build inside Cresta from day one.
HTML · Nginx · EC2 · SSL
"The best GTM hire isn't the person who knows the most tools. It's the person who builds the system that makes the tools irrelevant."
— The thing I'd say to Ping Wu in the first 30 minutes
The stack — what powers this
Temporal Cloud
Durable workflow orchestration — enrichment pipeline, Mystery Shopper runs, memory management
$100–300/mo depending on activity volume
Vapi
Outbound AI phone calls — IVR navigation, transcript extraction, platform fingerprinting
$0.10/call · usage-based
TheirStack
174M job posting database — CCaaS stack detection, VP CX hires, migration signals
$59/mo
Grok (xAI)
Real-time X/Twitter + web signals — CX leader intent, competitor mentions, IVR complaints
$50/mo
People Data Labs
Contact enrichment — title, tenure, LinkedIn profile, email verification
Usage-based · ~$0.04/record
FullEnrich
Email + phone number from LinkedIn URL — final enrichment step before sequence launch
Usage-based · ~$0.10/contact
SEC EDGAR
10-K/10-Q/8-K risk factor extraction — agent attrition, CX transformation budget signals
Free
Claude (Anthropic)
Transcript analysis, sequence generation, voice profiling, signal enrichment — LLM backbone
~$15–40/mo at this volume
context.dev
Brand API — colors, logos, fonts, company description from domain name
API key · low volume
📊
Forrester Wave: Conversational AI 2024
Cresta named Leader — highest Current Offering score in the category. Referenced in Rock 01 and sequences.
Forrester.com →
✈️
United Airlines Case Study
15% AHT reduction, 15% wait time reduction, 97% agent satisfaction. Used as sequence anchor in every airline and telecom touch.
Cresta.ai/customers →
📡
Optimum (Altice USA) Case Study
Existing customer — Telecom vertical win used as peer reference in Comcast and T-Mobile sequences.
Cresta.ai/customers →
🔍
LinkedIn Scraper — 461 posts
Playwright-based scraper with date extraction from URN binary encoding. Available on request.
scrape_rachit.py · see scraper code →
The point
This isn't a job application. It's a working system you can run on day one.
Every GTM engineer says they can build signal pipelines, run enrichment, and write sequences. Most of them mean they know what those words mean.

This portal is a running system. The Mystery Shopper has called 63 contact centers. The heat scores are computed from 8 live sources. The sequences open with what the AI found on the actual call. The voice system came from analyzing 461 posts.

If Cresta hires me, this is the infrastructure I'll run inside Cresta. The accounts will be Cresta's territory. The Mystery Shopper will dial real targets. The sequences will go to real CCOs. The only question is whether you want to start from here or from scratch.