Lead Generator

Overview

A personal applied-AI project: a company-intelligence pipeline that extracts website, funding, competitor, branding, marketing-copy and tech-stack signals.

Challenge

External data providers stall and fail — a useful intelligence report has to degrade gracefully instead of collapsing when one source does.

Results

Structured, machine-consumable company reports that survive provider outages — built for downstream product use.

My Ownership

• Built the pipeline extracting signals from 4+ external data sources
• Orchestrated parallel enrichment with per-call timeouts, exponential backoff and graceful degradation so reports still generate when providers stall or fail
• Validated LLM outputs with Zod schemas to produce structured JSON reports and reduce malformed-response failures

Technical Implementation

• Parallel fan-out with independent failure domains per provider — one dead source never kills the report
• Schema-validated LLM output (Zod) instead of trusting free-form JSON from the model

Engineering Challenges

LLM pipelines are distributed systems: timeouts, retries and validation matter more than the prompt.