An AI-powered B2B lead generation platform that transforms scattered intent signals into a curated pipeline of high-intent prospects, helping sales teams close 4x more deals.
Sales reps spend 40+ hours per week manually researching leads, sifting through news, filings, and job postings to find buying signals.
AI-only discovery loops produced unverifiable prospects, causing RevOps teams to lose trust in automated lead generation.
Intent signals were scattered across SEC filings, press releases, job boards, and news feeds with no unified view or scoring.
Account teams lacked defensible reasons to prioritize one company over another, leading to spreadsheet jockeying and inconsistent results.
A unified AI-powered platform that ingests, normalizes, scores, and surfaces B2B intent signals from multiple verified sources, all in one workspace.
Monitor funding announcements, hiring activities, and technology adoption across all B2B industries from verified sources.
Advanced algorithms evaluate signal strength, business impact, and buying stage with confidence levels and narrative rationale.
Identify companies actively seeking solutions with proprietary intent scoring (Hot/Warm/Cold segments).
Track signal trends, prospect quality, and conversion rates with detailed dashboards, charts, and funnels.
Get real-time answers on signals and prospects, fine-tuned with proprietary data and web search, directly inside your workflow.
One-click sync to Salesforce, HubSpot, Zapier, and Make with CSV/PDF export capabilities.
Turning complex problems into powerful features.
Challenge: Replacing brittle AI-only discovery with a deterministic pipeline that could ingest signals from SEC filings, press releases, and job boards.
Solution: Built modular adapter architecture with cursor tracking, dedupe guards, and replayable logs. Each signal carries provenance, timestamps, and match confidence.
Challenge: AI model outputs required validation, and ambiguous company matches led to false positives in the scoring pipeline.
Solution: Enforced Zod schema validation with retry logic. Composite scores include narrative rationale so users understand why a prospect is ranked.
Challenge: Sales demos failed when upstream data feeds experienced outages, damaging trust in the platform.
Solution: Implemented multi-tier caching: Supabase snapshots → Vercel blob storage → Redis → local filesystem → mock seeds. Zero downtime during rollout.
Challenge: Raw payloads from SEC, press RSS, and job boards had inconsistent formats, making unified analysis impossible.
Solution: Canonical signal normalization with company matching (domain/ticker + fuzzy aliases). Cached demographics enable instant lookups and filtering.
Challenge: Starting with real estate but needing architecture that could expand to any B2B vertical without code rewrites.
Solution: Universal signal detection where adapters read from database config. Adding new industries or feeds is a configuration change, not a deployment.
Challenge: AI assistants in similar products hallucinated leads and provided recommendations without evidence.
Solution: Xylem AI chat uses gateway-first fallback - structured answers from verified data before escalating to Generative AI. Every answer cites supporting signals.
Our target outcomes for delivering tangible value to sales teams and predictable revenue.
| Metric | Industry Standard | Our Target |
|---|---|---|
| Outreach Attempts | 1,000 | 200 |
| Response Rate | 2% | 15% |
| Conversion Rate | 5% | 20% |
| Research Time | 40 hrs | 12 hrs |
As a Founding Engineer, I built the full-stack system using Next.js 15 and Supabase/PostgreSQL, integrated Generative AI Pro for intelligent signal scoring, built the modular adapter pipeline for multi-source ingestion, and established TypeScript strict-mode standards that ensured zero ingest errors across thousands of signals.