AI & Automation Solutions Engineer

Natasha Weiner

I turn the operational work your team does by hand into production AI systems — typed, guardrailed, and running unattended.

NYC · Remote-based services

Recently Shipped

Proof over promises — systems and sites that actually run.

Personalized Budget App

LIVE
Live demo →

A personal-finance agent running in production, with unattended auth and guardrailed writes.

  • Python
  • launchd
  • Tailscale
  • Claude (headless)
  • SSE streaming
  • Vanilla JS
  • Production system that has run continuously on a Mac mini via launchd, reachable privately over Tailscale.
  • Unattended authentication: a self-healing login (credsync) keeps headless jobs authorized without a human in the loop.
  • Guardrailed writes: the read-only assistant is hard-locked to Read/Grep/Glob with Bash/Write/Edit disallowed — verified that write attempts are refused.
What does this do, in simple terms?

A private dashboard that watches money so nobody has to. Every day it reads incoming bank and card alert emails, keeps itself up to date, and shows one simple number: how much is safe to spend before the next bills pull.

It also answers plain-English questions — “can this month’s rent be covered?” — by reading the real numbers. And it’s built so the AI can see everything but can never change or move anything. That same pattern fits any business workflow that needs constant eyes on incoming data, without risk.

Etsy Agent

DAILY · AUTOMATED

A multi-agent content pipeline with typed contracts between every stage.

  • Python 3.11
  • FastAPI
  • APScheduler
  • Anthropic SDK
  • Pydantic
  • SQLite
  • WeasyPrint
  • Seven agents run in sequence; each takes one Pydantic model in and returns one out — agents never call each other, a scheduler orchestrates.
  • Typed contracts between stages make each step independently testable and the whole chain observable.
  • A SQLite feedback loop feeds measured performance (dead niches, untested niches, conversion by price band) back into the next run’s decision.
What does this do, in simple terms?

A robot shopkeeper. Once a day it checks what’s been selling in the shop, picks an idea for a new product people are actually searching for, designs it, writes the listing, posts it — then schedules a week of promotion on Pinterest.

Each step is a separate small worker that hands its output to the next one, so when something breaks it’s obvious exactly which step to fix. That’s the same way any dependable business pipeline should be built — and it learns from its own sales data, so it gets smarter every week.

More shipped work — client sites & smaller builds

Skin 911

A personal health dashboard that pulled years of scattered history, photos, and records into one page a doctor could actually scan during a visit.

Resume Parser

An architecture sample: parses resumes, embeds them, and ranks candidates against a role. A compact example of a retrieval + ranking service.

github.com/tashaweiner/resume-parser

Misch Implant Institute

Dental education platform

A 40-year dental-implant education institute. I built the Squarespace site that turns a deep course catalog into a clear path from first course to fellowship.

misch.com

Michele Stocknoff

Personal brand site

A full personal-brand site for a private practice — built on Squarespace to move visitors cleanly from “who she is” to booking a first conversation.

michelestocknoff.com

Working with me

I take fixed-scope contracts with operations-heavy teams and work the way the best technical companies deploy their own engineers — embedded. No pitch deck, no handoff to “the team”: I work inside your stack, find the workflow that hurts, and build the system that fixes it, until it runs without me.

  1. Scope
  2. Embed
  3. Ship
  4. Hand off

The full playbook →

About

  • I’m an engineer in New York. I’ve spent my career building data-heavy production systems — most recently healthcare cost-transparency infrastructure used by real patients.
  • I just wrapped Qwasar Silicon Valley’s 7-week Agentic AI course — no lectures, all building: agents for travel planning, customer support, and document search + Q&A, with hands-on RAG, multi-agent orchestration, and LLM fine-tuning.
  • Earlier training: an AWS Solutions Architect certification, Capital One’s Machine Learning bootcamp, and a graduate certificate in Bioinformatics.

I wrote about the cost-transparency work on Nayya’s blog: Cost Engine: How We Solved Healthcare’s Biggest Blind Spot nayya.com

Natasha Weiner

Skills

Languages

  • Python
  • TypeScript
  • JavaScript
  • Java
  • SQL

Backend & Web

  • FastAPI
  • Node.js
  • Express
  • React
  • Tailwind
  • GraphQL

Cloud & Infra

  • AWS (Lambda, S3, RDS)
  • Docker
  • Terraform
  • GitHub Actions
  • Redis
  • Kafka

AI & Agents

  • LangGraph
  • Multi-agent orchestration
  • RAG
  • LLM fine-tuning
  • AWS Bedrock
  • Hugging Face
  • MCP
  • Anthropic API
  • OpenAI API

Get in touch!

Best fit: operations-heavy teams with a specific, repetitive workflow that should be running itself. Tell me the workflow and what it's costing you.

Opens your email app — or write me directly at natashaweiner@gmail.com.