Full Stack Developer & AI Engineer
RIAX — Financial Advisor SaaS
AI-powered platform for financial advisors — meeting intelligence, behavioral profiling, SEC compliance tracking, and Kanban task management in one enterprise SaaS
Overview
RIAX is an enterprise SaaS platform built for Registered Investment Advisors (RIAs). It connects to Zoom and Teams via Recall.ai bots, transcribes meetings with AssemblyAI (PII-redacted), and feeds transcripts into a Claude Opus 4.5 AI pipeline on AWS Bedrock. The AI produces multi-length summaries, a Meeting Intelligence Score (MIS), behavioral profiling (e.g. 'Anxious Investor', 'Risk Averse', 'Cost Conscious'), SEC Rule 204-2 compliance flags, revenue opportunity detection with estimated values, and a behaviorally-adapted follow-up email draft. All outputs are persisted across 71+ PostgreSQL tables with full multi-tenancy via acts_as_tenant. A Kanban task engine with AI-scored priorities, a compliance audit trail with CSV/PDF export, and integrations with Salesforce, Redtail, Wealthbox, Orion, and Black Diamond complete the platform.
The Challenge
Financial advisors were manually writing meeting notes, missing compliance obligations under SEC Rule 204-2, and losing track of client follow-ups. The core challenge was building an AI pipeline that could reliably extract structured intelligence from unstructured meeting transcripts — without hallucinating financial data, violating compliance rules, or losing context across a multi-tenant system where every query must be company-scoped. The platform also needed to integrate with the fragmented ecosystem of CRMs and wealth management platforms advisors already use.
How We Solved It
Claude Opus AI Meeting Analysis Pipeline
Every meeting transcript flows through a single Claude Opus 4.5 invocation (AWS Bedrock, 20K token output, 10-min timeout) with behavioral classification definitions and live portfolio data (AUM × fee rate = estimated revenue) injected as structured prompt context. Claude returns a strict JSON schema: three summary lengths (250–300 / 500–750 / 1,000–1,500 words), MIS score, behavioral profile with confidence scores, compliance flag array, revenue opportunities with probability and estimated value, and a follow-up email draft — all persisted to separate DB models in a single transaction.
SEC Rule 204-2 Compliance Engine
A post-analysis ComplianceScanJob screens every AI-generated email for regulatory violations (guaranteed returns, insider information, misleading statements) using configurable regex-based compliance patterns with severity levels (Critical, High, Medium, Low) and auto-action rules. The Compliance Dashboard shows a 100-point score, flag statistics per advisor, violation trends, and a full audit trail exportable as CSV or PDF. Admins can create new patterns in simple mode (natural language → AI-generated regex) or advanced mode (direct regex input).
Behavioral Profiling & Client Intelligence
Behavioral definitions are stored in a Behavior DB table and injected dynamically into every AI prompt. Claude classifies each client against the taxonomy (primary + secondary behavior with confidence scores) and builds behavioral profiles over time. Client Health Scores aggregate meeting data, task completion, and behavioral signals into a 4-tier score (max 70 points). The platform tracks 3,297 clients needing attention alongside those in good standing.
Task Command Center with AI Scoring
A full Kanban + table task engine with AI-scored priorities. Tasks are sourced from CRM integrations (Salesforce, Redtail, Wealthbox) or created manually, each scored 0–100 by an AI priority model. The Kanban board supports drag-and-drop via @dnd-kit with configurable stages per board. A SyncToCrmJob pushes completed tasks back to source CRMs. 86 open tasks, 30 archived — all visible in filterable table and Kanban views.
Multi-Tenant Architecture Across All Layers
acts_as_tenant scopes every query across all 71+ tables by company_id — no cross-tenant data is possible at the ORM level. The Rails backend exposes 41 jsonapi-serializer serializers across auth, meetings, clients, tasks, compliance, sync, and AI suggestion namespaces. Devise + devise-jwt handles API auth, with TOTP MFA (rotp), magic-link login, Google/Microsoft OmniAuth, and Rack-Attack rate limiting.
Wealth Platform & Calendar Integrations
Full bidirectional sync with Orion and Black Diamond (accounts, positions, transactions, benchmarks, AUM snapshots) via NdexIngestionJob. Google Calendar and Microsoft Outlook sync for meeting scheduling. Recall.ai dispatches recording bots to live Zoom/Teams meetings. AssemblyAI handles transcription with PII redaction. Every AI call writes a UsageEvent with cost tracking — viewable in the Cost Tracker dashboard with daily rollups.
Authentication & Security
Main Dashboard
Task Command Center
Meeting AI Analysis
Compliance Dashboard
Notifications & Help Center
Results & Impact
Enterprise-Grade AI Pipeline
Claude Opus 4.5 via AWS Bedrock — 20K token output, 10-min inference timeout, structured JSON output with 6 distinct AI deliverables per meeting analysis.
71+ Tables, 200+ Indexes
186+ migrations, full multi-tenancy via acts_as_tenant across all tables — zero cross-tenant data leakage possible at ORM level.
100-Point Compliance Score
SEC Rule 204-2 compliance dashboard with configurable regex patterns, severity levels, auto-block rules, audit trail, and CSV/PDF export.
86 Active Tasks Tracked
AI-scored Kanban + table task engine with CRM sync (Salesforce, Redtail, Wealthbox), drag-and-drop board, and automated priority scoring.
8 Third-Party Integrations
Google Calendar, Microsoft Outlook, Salesforce, Redtail, Wealthbox, Orion, Black Diamond, Recall.ai — all synced bidirectionally via background jobs.
Full Auth Security Stack
Devise + JWT + TOTP MFA + magic-link login + Google/Microsoft OmniAuth + Rack-Attack rate limiting — 5-attempt MFA lockout threshold.
Technical Architecture
Have a similar project?
I'm open to freelance projects and full-time roles. Let's build something great together.