Nicholas Iazzetta
Building the tools to maximize efficiency
New to Contra
Nicholas is ready for their next project!
Automated B2B Lead Generation & Data Pipeline
Description:
Engineered a high-volume, automated outbound lead generation system for a B2B financial brokerage (ZedCapital Partners). The challenge was bypassing manual prospecting limits and maintaining domain reputation at scale.
I built a pipeline that programmatically scrapes and extracts target decision-maker data (CEOs/Founders in specific verticals), scrubs the raw datasets through a verification API to eliminate catch-all/risky emails, and automatically routes the clean data into sequenced email campaigns. The system safely paces sending volume to maximize deliverability and avoid spam traps.
Impact:
Replaced manual prospecting with an automated scraping pipeline.
Integrated multi-step API workflows (Extraction -> Verification -> Delivery).
Maintained 99%+ deliverability by engineering strict data-cleaning rules.
1
15
Gredia: Allergy Scanner and Affiliate Recommendation Engine
The Problem:
Consumers with severe food allergies struggle to find safe alternatives to products they love, while manually reading complex ingredient labels is time-consuming and error-prone.
The Solution:
Gredia is a B2C application that scans food products, instantly filters out dangerous allergens using a proprietary rule engine, and recommends safe, highly-matched alternative products via Amazon's PA-API to generate affiliate revenue.
Technical Highlights:
Massive Data Normalization: Engineered a Node.js script using the OpenAI Batch API to process and normalize unstructured ingredient data for over 3,500,000 products.
High-Performance Search: Replaced expensive vector searches with a heavily optimized PostgreSQL Full Text Search (FTS) and strict Regex filtering pipeline in Supabase, dropping latency to milliseconds.
Resilient API Integrations: Built a custom caching layer (Stale-While-Revalidate) to respect Amazon PA-API's strict 1-request-per-second limits while maintaining instant UI load times.
0
15
Gredia Case Study: The Scope of US Food Toxicity
Before engineering the Gredia filtering engine, I needed to quantify the actual scale of the problem in the US food supply. I extracted and mapped the ingredient data for 447,000 USDA food products to visualize exactly how many products are contaminated with the 17 most common toxic ingredients and additives.
Visualization 1 (Red/Green): The Total Landscape (Nomic Atlas)
This map clusters all 447,000 products based on their ingredient profiles. The color coding exposes the severity of the toxicity score (the number of the 17 harmful ingredients present). It visually confirmed the thesis: the majority of the US food supply is heavily processed and tightly clustered around toxic additives.
Link: https://atlas.nomic.ai/data/niazzetta719/gredia-the-grocery-store-map-full-scale/map/cdb38e19-e9e9-466a-b1fc-8ac5425e4238#gVVn
Visualization 2 (Yellow/Blue): The Gravity of Toxicity (Cosmograph)
I isolated the 120,000 products that contained at least one of the targeted toxic ingredients. This force-directed map applies gravity to pull each food product toward its parent toxins. The color gradient highlights the density of contamination, specifically exposing the dense red cluster in the dead center—highly processed products containing 6 or more toxic additives simultaneously.
Link:https://run.cosmograph.app/public/33a71bb2-5250-4c6d-8277-d3652f4012f7
0
20
The Problem:
Lenders and brokers in the Merchant Cash Advance (MCA) space waste thousands of hours manually extracting Average Daily Balances (ADB) and gross deposits from unstructured, messy bank statement PDFs.
The Solution:
Macram is a high-throughput AI SaaS that automates this entire process. It ingests raw bank documents and runs them through a custom 10-stage LLM pipeline to instantly output a flawless financial underwriting matrix.
Technical Highlights:
Advanced LLM Orchestration: Built a multi-model pipeline using Anthropic's Claude Haiku for fast data chunking/minification and Sonnet for complex synthesis.
Deterministic Output: Engineered a state machine that forces unstructured OCR text into strictly typed JSON fragments and accurate financial calculations.
High-Concurrency Architecture: Designed a custom concurrency "Valve" to process heavy document loads simultaneously without triggering provider rate-limit crashes.
0
21