- AI / LLM
- AWS
- Javascript
- MySQL
- PHP
Aimee: AI Medical Coding Assistant
Find-A-Code's AI research assistant that gives medical coders fast, audit-ready answers grounded in trusted coding data instead of the open web.
Overview
Aimee is the AI research assistant built into Find-A-Code, a medical coding reference used every day by coders, billers, and auditors. Ask about ICD-10, CPT, HCPCS, modifiers, NCCI edits, Medicare coverage, fees, or payer policies, and Aimee researches the question the way an expert coder would: it looks up the actual code data and policy sources, then replies with a concise, audit-ready answer. When a question is ambiguous, it asks a clarifying question instead of guessing.
What started as a chatbot prototype has grown into a standalone AI service that powers Aimee on Find-A-Code, a sibling coding product, and an authenticated API for partner integrations.
My role
I helped build Aimee’s first research tools in 2025 and have led the engine end to end since early 2026: the agent loop, retrieval, the asynchronous job pipeline, prompt management, patient-privacy safeguards, usage-based billing, and the move from the main web application to its own service.
Highlights
- A tool-using AI agent grounded in real data. About 50 purpose-built tools let the model query code sets, NCCI edits, Medicare coverage, fees and RVUs, crosswalks, and payer policies directly, so answers come from authoritative data rather than model memory.
- Retrieval over reference libraries. Semantic search with pgvector embeddings and hypothetical-answer (HyDE) queries, with access checked against each user’s subscription on every tool call.
- Answers you can trust. Every link and code reference is validated before it reaches the user, and a growing suite of regression tests built from real-world edge cases keeps the guardrails from slipping.
- A multi-provider LLM layer. Provider-agnostic adapters with automatic failover, per-request cost tracking, prompt caching, and a Deep Dive mode that sends hard questions to a stronger reasoning tier.
- Scalable, streaming architecture. Questions run as jobs on AWS SQS across a pool of workers, with Redis-backed live streaming, job leases, and automatic recovery of stalled work.
- Patient-privacy safeguards. Direct patient identifiers are detected as users type, flagged in real time, and automatically redacted from stored conversations.
- A prompt platform. Versioned prompts with question routing, simulation, and draft, publish, and audit workflows so the team can improve answers safely.
- A zero-downtime service split. Extracted Aimee into its own service with a versioned REST API, ran both engines side by side, and moved traffic over gradually.
- Usage-based billing. Monthly response quotas, packages, and top-ups, plus API keys with per-client scopes and limits.
Features for coders
- Streaming chat with conversation history, printing, and PDF export
- Deep Dive mode for complex questions
- Attach a document or URL to add context to a question
- Voice input
- Inline code tables and links back to Find-A-Code
- Thumbs up and down feedback that drives continuous improvement
Tech
PHP 8, MySQL, PostgreSQL with pgvector, Redis, AWS (SQS, EC2, RDS), a vanilla JavaScript chat client with live streaming, and multiple LLM providers behind a common interface.