nweat

Hey, I am Nicole W.

AI Product Engineer

Automation & AI | Full-Stack & Scalable Systems

React & Vite TypeScript Python LangChain LangGraph Vector Search RAG OpenAI / Claude Cursor Agent Skills English Writing GCP & AWS Docker

AI Projects

Production

AI Release Notes Generator

FastAPI service that drafts editor-ready release notes when a manufacturing guide is released - comparing the draft to the latest published revision and generating structured notes with Claude Haiku via Bedrock.

Schema-constrained JSON
LangSmith managed prompts for iteration

FastAPI Bedrock Claude Haiku Structured Outputs NestJS AWS ECS Cost-aware prompts Datadog

AI engineering toolkit

Engineering Brag Pack Skill

A reusable agent skill that turns real work evidence into a performance-review talk track. Built as part of an AI engineering toolkit.

Pulls from Jira, GitHub, Notion (+ optional peer shoutouts) via MCP
Human confirm-gate so noise never becomes false brags
Impact + Signal + Story format → interactive Canvas for 1:1s

Agent Skills MCP Cursor Canvas Prompt Design

Freelance · Eval-first

RAG Chatbot Prototypes & Evals

Scoped RAG FAQ/SOP chatbots with an evaluation harness

Golden question sets with expected citations / “I don’t know” cases
Score retrieval hit-rate and answer faithfulness before shipping prompt tweaks
Offline eval scripts + traces so regressions show up before customers do

RAG Evals Golden Fixtures Embeddings

How I build with AI

Cursor isn’t autocomplete for me, it’s an orchestrated workflow

Daily driver

Agent Skills

I develop against production-grade agent skills so coding agents follow senior engineering habits across every phase- define, plan, build, verify, ship.

Skill routing: pick the right workflow per task
Non-negotiables: surface assumptions, manage confusion
Verification is required — passing tests / runtime evidence
MCPs for Jira, GitHub, Notion, and AWS when the task needs live systems context

addyosmani/agent-skills →

Agent Skills Spec-driven TDD Code Review Cursor Rules

Architecture Diagrams

LLM RAG Pipeline with Vector Search

Architecture for reducing LLM context costs by retrieving relevant documents via vector similarity search with metadata filters before prompt assembly.

User Question
     │
     ▼
┌─────────────┐     ┌──────────────────┐
│  Embed query │────▶│  Vector search   │
└─────────────┘     │  + metadata      │
                    │  filters         │
                    └────────┬─────────┘
                             │ top-k chunks
                             ▼
                    ┌──────────────────┐
                    │ Prompt assembly  │
                    │ (system + ctx)   │
                    └────────┬─────────┘
                             ▼
                    ┌──────────────────┐
                    │ LLM (Claude /    │
                    │ OpenAI / Bedrock)│
                    └────────┬─────────┘
                             ▼
                        Grounded answer

Vector DB Metadata Filters LLM API

AI Release Notes Generator

Compare draft vs latest released revision, summarize changes, and return editor-ready notes via Claude Haiku on Bedrock. The API is designed in way to be registered as a tool via MCP for the AI Assistant.

User request
         │ 
         ▼
   ┌───────────┐      ┌──────────────────┐
   │ proxy     │─────▶│ release-notes-api│
   │           │      │  (FastAPI / ECS) │
   └───────────┘      └────────┬─────────┘
                               │ 
                               ▼
                      ┌──────────────────┐
                      │ guides           │
                      │ draft + release  │
                      └────────┬─────────┘
                               │ diff summary
                               ▼
                      ┌──────────────────┐
                      │ Bedrock Haiku    │
                      │ (schema JSON)    │
                      └────────┬─────────┘
                               ▼
                   Editor-ready release notes

FastAPI Bedrock Claude Haiku ECS