Senior Node.js Engineer for Secure Operations Dashboard and Local AI Call-Quality Pipeline

Less than 30 hrs/week We are seeking a senior backend/full-stack engineer to help strengthen and scale an existing private operations dashboard for a service business. The application is already in production and includes: - A Node.js and JavaScript backend - A browser-based management interface - Private GitHub source control - Railway hosting - Field-service platform integrations - Local speech-to-text processing - A local language model that evaluates recorded customer-service calls - An AI coaching workflow that uses verified call findings The call-quality system must eventually process approximately 400-500 recorded calls per business day. Results should be ready by the end of the business day and include accurate employee attribution, role-specific grading, supporting transcript evidence, and safe handling of uncertain results. This is an established system with substantial working code. We are not looking for a ground-up rewrite. The successful developer will inspect the current implementation, preserve working safeguards, identify bottlenecks, and make small, reviewable improvements. INITIAL PAID TRIAL The selected developer will begin with a paid trial of approximately 10-20 hours. The trial will include: - Review the relevant architecture, queues, analyzer, storage, benchmark package, and coaching integration. - Identify the most important accuracy, throughput, reliability, maintainability, and security risks. - Evaluate whether the existing benchmark can measure false accusations, missed behaviors, employee attribution, abstention, processing time, and duplicate processing. - Produce a prioritized implementation plan that clearly separates measured facts from assumptions. - Complete one mutually approved, isolated improvement through a small GitHub pull request. - Include focused tests, documentation, operational risks, and rollback instructions. The trial does not include production deployment, unrestricted production access, or a broad rewrite. If the trial is successful, the engagement may continue through additional bounded assignments covering analyzer throughput, grading accuracy, queue reliability, coaching integration, dashboard improvements, and general site maintenance. REQUIRED EXPERIENCE - Strong Node.js and JavaScript experience - Production queue and background-worker design - Idempotency, retry handling, caching, and immutable result history - Structured-output language-model pipelines - Speech-transcript or call-analysis systems - Evidence-grounded classification and safe abstention - GitHub pull-request workflows and protected branches - Secure handling of credentials and private customer or employee data - Performance measurement and capacity planning - Clear written technical communication HELPFUL EXPERIENCE - Railway - ServiceTitan or similar field-service platforms - Local language models - llama.cpp or comparable local inference runtimes - Qwen or similar structured-output models - Whisper or comparable speech-to-text systems - Speaker diarization - AI evaluation harnesses - Employee coaching or quality-assurance systems SECURITY AND WORKING RULES - All work must occur in an assigned branch through small pull requests. - The contractor cannot merge or deploy their own work. - Production credentials, environment variables, customer recordings, employee HR information, billing accounts, and domain controls will not be provided unless separately authorized for a specific need. - Initial testing will use sanitized or specifically approved data. - Proprietary code, transcripts, recordings, employee data, customer data, and business information may not be uploaded to unapproved AI services. - Any AI coding tools used by the contractor must be disclosed. - AI-generated code must be understood, tested, and reviewed by the contractor before submission. - No subcontracting or sharing access without written approval. - No broad rewrites, force pushes, destructive repository operations, or unapproved dependency changes. - Sensitive central application files must be patched carefully rather than replaced wholesale. - Production integrations remain read-only unless a separately approved task explicitly requires otherwise. - Every change must include enough information to review its purpose, affected files, verification, risks, and rollback. - All work is subject to owner and independent technical review. - The selected contractor will be required to follow confidentiality, intellectual-property, data-handling, and access-removal requirements. ACCURACY REQUIREMENTS The call analyzer affects employee grades and coaching. Accuracy and traceability are more important than producing a result for every call. The system must: - Never invent transcript evidence. - Never assign uncertain speech to an employee. - Never treat missing evidence as employee failure. - Preserve independent checks for consequential negative findings. - Leave unresolved findings unscored. - Prevent duplicate processing from inflating grades. - Link every coaching observation to verified employee-specific evidence. - Preserve prior results and revision history. - Clearly report discovered, pending, failed, uncertain, unscored, and verified coverage. - Keep daily processing separate from development experiments. WHAT SUCCESS LOOKS LIKE - Changes are small enough to review and reverse. - Existing functionality and audit history are preserved. - No unsupported negative employee finding becomes a grade. - Duplicate reviews do not contribute multiple times to an employee score. - Each coaching observation can be traced to verified evidence. - Throughput and accuracy claims are supported by measurements. - The system progresses toward same-day processing of 400-500 calls. - Daily operations do not depend on a continuously active development agent. - Production remains protected throughout development. HOW TO APPLY Please answer all six screening questions directly. State your hourly rate and estimate how many hours you would need for the initial trial. You may also provide a fixed-price alternative for the trial. Include one or two examples of relevant systems you personally built or materially improved. Explain your exact contribution, the scale involved, and how you protected production data and deployments. Generic agency introductions, undisclosed subcontracting, and proposals that recommend rebuilding the entire application before inspecting the existing system will not be considered. Cost matters, but selection will be based on relevant experience, security judgment, communication, and demonstrated ability - not the lowest rate alone.