ZZJL-001
FILE · 00 — INDEXCLASSIFICATION · PUBLIC
DOSSIER ZJL-001 · OPENED1–2 WK SPRINTS · CHICAGO

ZacharyJohn·Lavengco.

AI full-stack engineer shipping generative systems, data pipelines, and interfaces that feel alive.

For hire · I help Chicago teams fill dev gaps in automation, data, and AI integration — when hiring full-time is too slow.

SPECIMENLIVE
Zachary John Lavengco
ZJL-001
ID
EN-US
LANG
CHI/CST
TZ
EXPERIENCE
6+ YRS
RAG · SEARCH TIME
−85%
HACKATHON WIN
BEST AI · 2024
LIVE DEMOS
8+ (200+ PPL)
★ TRUSTED BYLAST SHIPPED AT —
CurrentNorthern Trust
PastWalsh Group
EducatedUIC Honors College
CertifiedMicrosoft
CertifiedDatabricks
§ 01Offerings · For Hire

What I ship for teams.

Three plain-named lanes I plug into when full-time hiring is too slow.

Pitch · I help Chicago teams fill dev gaps in automation, data, and AI integration — when hiring full-time is too slow.

SERVICE · 011–2 weeks

Automation.

Take the manual work off your team's plate.

OCR + AI extraction, document parsing, internal tools, workflow bots, scrapers, integrations. The boring repetitive stuff — done once, automated forever.

  • Salesforce + Tesseract OCR pipeline · cut document work by 95%
  • Internal web tools across an enterprise org
  • API integrations + scheduled jobs
Save 80%+ ops timescope this →
SERVICE · 021–3 weeks

Data.

Get your data from messy to useful.

Pipelines, dashboards, cleanup, modeling, medallion architecture, vector DBs. Make decisions on numbers you actually trust.

  • Production-grade Databricks pipeline at Northern Trust
  • Postgres modeling + Liquibase migrations
  • SQL audits, dashboards & reporting
Decisions you can trustscope this →
SERVICE · 031–3 weeks

AI Integration.

Production AI — not demos.

RAG chatbots, agent orchestration, prompt evaluations, custom LLM workflows, MCP tooling. The AI that actually moves a business metric.

  • Northern Trust's first production GenAI app · search time −85%
  • LLM-as-a-judge evaluation framework
  • Agent + tooling stacks (LangGraph / MCP)
Real production AIscope this →
Brief budgetsAsync-friendlyNDA-friendlyHourly or fixed-fee
PROCESS · HOW WE'LL WORK

Four steps. No surprises.

Book the 15-min
  1. STEP · 01/BRIEF
    Brief.

    Free 15-min intro. Send the rough idea — I come back with a shape.

  2. STEP · 02/SCOPE
    Scope.

    1-page proposal: outcome, timeline, fixed-fee or hourly. No surprises.

  3. STEP · 03/BUILD
    Build.

    Async, daily updates, code in your repo from day one. NDA-friendly.

  4. STEP · 04/SHIP
    Ship.

    Hand-off plus two weeks of post-launch support. Docs included.

§ 02Voices · Field Notes

On the record.

What two people say after working with me — out loud, on their own LinkedIn.

FIELD REPORT · 01VERIFIED
Zach treats your business outcome as the spec, not the ticket. We went from a fuzzy idea to a working tool in a few weeks — and our team has been using it every day since. I'd hire him again before I'd post a job.
Michael Manos
FIELD REPORT · 02VERIFIED
Zach is who I call when something has to ship and the timeline is uncomfortable. Calm under pressure, asks the right questions early, and the work is production-grade — not a demo. Rare combination, in my experience.
Paul E. Prusakowski
Paul E. PrusakowskiChief Strategist & Executive Producerverify on linkedin ↗
§ 03About / Manifesto

The Engineer, Declassified.

Raised in stacks, shipped in sprints, now living where models meet interfaces.

MANIFESTO
I write code that thinks. Bridging backend intelligence and front-end clarity — shipping production AI at a 135-year-old bank and side quests that won hackathons along the way.
01/Who

AI Full-Stack Engineer with 6+ years turning Python services and React interfaces into systems people actually use.

02/What

Architected a production RAG chatbot for wealth advisors at Northern Trust — cutting document search time by 85%.

03/How

Microservices on Azure, Databricks medallion pipelines, evaluation frameworks for LLMs, and a lot of prompt archaeology.

04/Why

Because the interesting work happens at the seam between model, data, and interface — and I like living there.

§ 04Chronicle

A log of production systems.

Six years, three stacks, one throughline: ship calm, useful software.

    • Contributed to the production launch of the company's first generative AI application on multiple platforms using RAG architecture on Azure, deploying function-app microservices that cut document search time by 85% across core advisor workflows.
    • Architected and coded the application's first production-grade Databricks data pipeline in a microservices architecture, ingesting from two internal data sources and designed to scale to future datasets.
    • Independently designed and built an AI testing framework for comparing AI configuration changes, benchmarking prompt efficiency and model performance across iterations.
    • Designed and implemented calculated and AI-specific evaluation metrics for LLM responses — including LLM-as-a-judge analysis — to measure response quality, hallucination rates, and model behavior.
    • Collaborated across agile/scrum teams and business users to ship production-grade data products in Databricks that feed a downstream AI Research Agent.
    • Presented and demoed the generative AI platform 8+ times to business and technical audiences (up to 200+ attendees), translating AI concepts into the bank's long-term AI vision.
    PythonAzure FunctionsRAGLangGraphDatabricksVector DBLLM-as-a-JudgeReactTypeScript
§ 05Capabilities

Toolkit, mapped.

A working inventory — languages, systems, and AI tooling I reach for daily.

CAPABILITY · 0616 ENTRIES

AI & ML

RAGPrompt EngineeringAgent OrchestrationLangGraphLLM-as-a-JudgeAI MetricsGenerative AIMCPPyTorchScikit-LearnHuggingFaceNLTKChatGPTClaudeCopilotCodex
LAST CALIBRATED · 2026SELF-REPORTED
§ 06Artifacts

Selected work.

Two artifacts: a production AI tool that won our hackathon, and a developer-first chat app.

Sep 2024
Intelligent Document Automation preview
Hackathon Winner · Best AI Solution
/01ARTIFACT · 01

Intelligent Document Automation

AI-Powered Document Parsing & Generation · Salesforce

An AI-driven document automation tool that scans withdrawal and contribution forms, detects fields, tables, and checkboxes, and regenerates updated digital documents — winner of the internal hackathon.

  • Scans scanned withdrawal and contribution forms using Tesseract OCR and OpenCV field detection.
  • Extracts metadata for each field and maps it to Salesforce data sources plus user inputs to auto-populate documents.
  • Reduced manual document creation time by ~95% for operations teams.
SalesforceTesseract OCROpenCVAI / MLPythonMetadata Mapping
2024
DevChat preview
/02ARTIFACT · 02

DevChat

Terminal-themed chat app built for developers

A chat application disguised as a developer's terminal — channels, code snippets, and real-time messaging wrapped in an IDE-like shell.

  • Built with Next.js, styled as a working developer terminal with channel sidebar and snippet-aware messages.
  • Google OAuth for authentication, Firebase Realtime Database for persistence and live sync.
  • Third-party API integrations and keyboard-first UX patterns.
Next.jsFirebaseGoogle OAuthREST APIsTypeScript
§ 07Credentials

Paper trail.

Certifications and a CS degree that keep the theory honest.

CERTIFICATIONS03 ITEMS
  • AZ-900CERT

    Microsoft Azure Fundamentals

    Microsoft
  • COURSECERT

    Generative AI Engineering with Databricks

    Databricks
  • COURSECERT

    Data Engineering with Databricks

    Databricks
EDUCATIONUIC
HONORS COLLEGE

University of Illinois at Chicago · Honors College

Bachelor of Science, Computer Science

Chicago, IL
B.S.
DEGREE
CS
MAJOR
CHI
CAMPUS
§ 08FAQ · Common Questions

Before we hop on a call.

The questions buyers usually ask in the first DM. Quick answers below.

  1. A · 01

    Usually within a week. Most engagements kick off with a 1–2 week scoped pilot so you can see velocity before committing further.