Logo Lanfrica

Adrxn2000/financeIQ

Domaine:

socioeconomic

Type de record:

software
Créateur:
Adr
Hôte:
FinanceIQ pairs a deterministic, logistic-style risk scorer with a Gemini-powered conversational interface. Users can chat with the assistant about portfolio risk, assess new loan applicants, and explore a dashboard of 100 synthetic applicants — all in South African Rand (R). # FinanceIQ — AI Loan Risk Assistant An AI-powered loan risk assessment chatbot for the South African retail banking sector. Built for the **Clickatell × CAPACITI AI Bootcamp 2026 — Week 4**. FinanceIQ pairs a deterministic, logistic-style risk scorer with a Gemini-powered conversational interface. Users can chat with the assistant about portfolio risk, assess new loan applicants, and explore a dashboard of 100 synthetic applicants — all in South African Rand (R). --- ## Features - **AI Chat** — Conversational assistant grounded in live portfolio data (powered by Gemini via the Lovable AI Gateway). - **Risk Assessment** — Submit a new applicant and receive a risk level, default probability, recommendation, key risk factors, conditions, and plain-English explanation. - **Portfolio Dashboard** — 100 pre-loaded applicants with sortable table, filter chips, donut chart, feature importance bar, and insight cards. - **Deterministic Scoring** — Reproducible risk scoring formula so results stay stable across requests. - **Rand Throughout** — All currency formatted as `R` (en-ZA). --- ## Tech Stack | Layer | Choice | |--------------|------------------------------------------------------------------------| | Frontend | React 19 + TanStack Start + Tailwind CSS (dark theme) | | Server | TanStack `createServerFn` handlers (TypeScript on Cloudflare Workers) | | AI | Google `gemini-3-flash-preview` via the Lovable AI Gateway | | Data | 100 synthetic applicants generated in-memory with a seeded RNG | | Tooling | Vite 7, TypeScript (strict), Bun | > **Note on the original spec.** The original brief proposed a Python/FastAPI + scikit-learn backend. Lovable runs on Cloudflare Workers (TypeScript only), so the same logistic-style risk model was ported to TypeScript server functions. Behavior matches …