Speed
Fintech
Data Collection Template — Daily Recon (MMR)
Daily reconciliation and anomaly-detection dashboard for Speed's customer & treasury funds
THE CHALLENGE
What problem existed before this app?
Speed holds customer funds and its own treasury across many sources — Speed nodes, exchanges, wallets, hardware wallets — in multiple assets (USDT, XAUT, USD, etc.). Every day the team needed to know: does the real money on hand still match what we owe customers, and what changed since yesterday? Answering that meant stitching together spreadsheets, live FX rates, and tribal knowledge, with no consistent trail of why a number moved.
Who was affected, and how?
Finance/treasury operators (daily reconciliation), engineering leadership (visibility into coverage), and auditors/reviewers who needed a defensible record of anomalies and explanations.
Previous process / workaround
Manual spreadsheet uploads, ad-hoc FX conversions using stale rates baked into the file, Slack threads
Cost of the problem
Several hours per day spent reconciling and re-computing USD values with fresh FX • Silent drift because spreadsheet USD columns used upload-time rates • Anomalies (10%+ KPI moves) noticed late or forgotten by the next day • No audit trail linking a number change to a human explanation
The solution
The Solution
Daily Recon ingests the team's daily reconciliation report, parses per-source per-asset native balances, and re-computes every USD figure from live CoinGecko rates — so Customer Funds, Speed Funds, and Total Holdings always reflect current market value, not the spreadsheet's upload-time FX. It surfaces day-over-day movement across KPIs, sources, and assets, flags anomalies over a configurable threshold, and lets staff annotate drastic changes so the "why" lives next to the number. An AI Fund Analyst runs on each new report, asks targeted questions about unexplained movements (respecting funds-in-transit / settlement-timing rules), and captures the team's answers as a permanent Q&A thread per day.
Top key features
Integrations & tech stack
The platform is built on a modern React + TypeScript frontend, a Lovable Cloud backend with Postgres and RLS, and edge functions for secure, advisory-locked operations.
Frontend
Lovable Cloud (Postgres + RLS, Auth, Edge Functions, Storage)
Backend
Lovable Cloud (Postgres + RLS, Auth, Edge Functions, Storage)
Integrations
Lovable Cloud (Postgres + RLS, Auth, Edge Functions, Storage)
Anything unique or clever
Server-authoritative metrics.
the AI is not allowed to author dollar figures. The edge function computes exact per-day, per-asset deltas and overwrites the model's metric_context, dropping any question that doesn't match a real computed movement — so the analyst can't hallucinate a number.
Not Available
Manual spreadsheet uploads, ad-hoc FX conversions using stale rates baked into the file, Slack threads
Not Available
Roles live in a dedicated user_roles table behind a SECURITY DEFINER has_role() function to prevent privilege-escalation via profile edits.
Results & Impac
Not Available