Overview We are an Australian hair colour company looking for an experienced full-stack developer to build a platform that automatically generates personalised hair colour recommendation emails using the Anthropic Claude API. We have a working system prompt (detailed AI instruction set) that has been thoroughly tested and validated. The system prompt instructs Claude to analyse a customer’s questionnaire responses and uploaded hair photo, then generate a personalised email recommending the right products, developer, application technique, and care products. We need a developer to build the integration layer that connects our existing tools and makes this work end-to-end in production. The Flow 1. A customer completes a questionnaire on our website (powered by LeadsHook), which captures their hair details, goals, and a photo of their hair. 2. The questionnaire data and photo are sent to a custom platform (the part you will build). 3. Your platform sends the questionnaire data + photo + our system prompt to the Claude API (Anthropic). 4. Claude returns a JSON response containing the email recommendation, photo assessment, and a flag indicating whether human review is needed. 5. Your platform parses the response and creates the email in Help Scout (our customer support tool): • If the AI flags it for human review → create as a draft with an internal note • If no review needed → create as a draft (initially; auto-send once validated) What We Provide • A complete, tested system prompt (plain text file, approx. 6,500 words) that contains all product data, decision logic, conditional content blocks, and reference examples • A developer handover document with full technical specifications, code examples (Python and Node.js), API setup instructions, error handling, and testing plan • A feedback guide explaining how we update the system prompt over time (you do not need to build a prompt editor — we update a text file and you deploy it) • Access to our LeadsHook, Help Scout, and Anthropic API accounts • 30+ historical questionnaire/email pairs for testing and validation What You Will Build • A middleware platform that receives questionnaire data + photo URL from LeadsHook (via webhook) • Fetches the customer photo from the URL and encodes it as base64 • Constructs the Claude API request with the system prompt + questionnaire data + photo • Sends the request to the Anthropic Claude API and receives the JSON response • Parses the JSON response and extracts the email body, subject, review flag, and metadata • Creates the email in Help Scout via their API (as a draft, with internal notes if flagged for review) • Logs every request/response for monitoring and debugging (timestamp, customer ID, recommendation summary, token usage, review flag) • Basic error handling: retry logic for API failures, fallback for invalid JSON, timeout handling Important Technical Details The system prompt MUST be loaded as the ‘system’ parameter in the Claude API call — not in the messages array. This is critical for consistent results. The complete system prompt text file must be sent in full with every API call. Do not truncate, summarise, or split it. The system prompt is approximately 6,500 words and is well within Claude’s 200K context window. The customer photo must be sent as a base64-encoded image within the user message content block, alongside the questionnaire data as plain text. Both must be in the same user message. The email body returned by Claude contains product names as hyperlinked text (not raw URLs). Your platform must preserve these hyperlinks when creating the email in Help Scout. Prompt caching should be enabled for cost optimisation — the system prompt is identical for every request, so caching reduces input token costs by 90%. Ongoing Maintenance We will update the system prompt text file periodically as we refine the AI’s recommendations based on feedback. Your platform should load the system prompt from a file at runtime (not hardcoded), so that we can update it without code changes. We will provide the updated text file and you replace it in the deployment. Essential Requirements Technical Skills • Proven experience with the Anthropic Claude API (or similar LLM APIs such as OpenAI). Must understand the difference between the system parameter and the messages array, and how to send images via the Vision API. • Strong backend development skills in Python or Node.js (our handover document includes code examples in both). • Experience building webhook-based integrations (receiving data from external platforms via HTTP POST). • Experience with REST API integrations (Help Scout API for email creation, LeadsHook webhook for data ingestion). • Understanding of base64 image encoding and handling image data in API requests. • Experience with error handling, retry logic, and logging in production systems. • Familiarity with prompt caching or willingness to implement it per Anthropic’s documentation. Desirable Skills • Experience with Help Scout API specifically. • Experience with LeadsHook or similar form/survey platforms. • Experience deploying and maintaining lightweight middleware/microservices (e.g., on AWS Lambda, Railway, Render, or similar). • Familiarity with HTML email formatting and preserving hyperlinks in email bodies. Working Style • Clear communication — we are a small, non-technical team and need things explained in plain English. • Willingness to work from our detailed handover documentation rather than reinventing the architecture. • Availability for a short testing and validation phase (2–4 weeks of parallel running where all emails are created as drafts for our team to review before we enable auto-send). • Based in a timezone compatible with AEST (Australian Eastern Standard Time) or willing to overlap for communication. Budget and Timeline • Budget: Open to proposals. Please provide a fixed-price quote for the initial build, and a separate hourly/monthly rate for ongoing maintenance. • Timeline: We would like the initial build completed within 2 weeks, followed by a 2 week validation period. • Ongoing: Minimal maintenance expected — primarily deploying updated system prompt files and occasional bug fixes. We do not anticipate frequent code changes. How to Apply Please include in your proposal: 1. A brief description of your experience with LLM APIs (Claude, OpenAI, or similar), including any projects where you integrated an LLM with external systems. 2. Your preferred tech stack for this project (language, hosting platform, etc.). 3. A fixed-price quote for the initial build and your hourly rate for ongoing maintenance. 4. Your estimated timeline for delivery. 5. Your timezone and availability for communication during AEST business hours.