New Way — AI Financial Intelligence & Revenue Forecasting — Case Study

01 // Context & Purpose

New Way is an AI-augmented Business Intelligence SaaS designed to eliminate manual spreadsheet analysis. Built with Next.js 16, React 19, and the Vercel AI SDK, users simply drag-and-drop raw CSV/Excel transaction datasets. The system automatically infers data schemas, renders interactive KPI cards, generates 30-day linear regression revenue forecasts, and deploys Google Gemini to analyze trends, explain hidden anomalies, and answer natural-language business questions in real time.

02 // The Engineering Challenge

Small businesses and financial operators often lack the budget for complex enterprise BI suites like Tableau or PowerBI, yet traditional spreadsheets cannot automatically explain underlying revenue drivers or forecast future trends without manual statistical modeling.

03 // Architecture & Implementation

Constructed a serverless, ultra-fast analytics architecture utilizing PapaParse for client-side streaming dataset ingestion, Recharts for fluid data visualization, local linear regression engines for 30-day predictive forecasting, and Gemini Flash streaming endpoints for context-aware financial reasoning.

Key Technical Deliverables:

✓ Built drag-and-drop CSV/Excel ingestion pipeline with automated column detection and data type inference.
✓ Implemented dynamic interactive financial dashboard with automated KPI calculations and trend visualizations.
✓ Engineered mathematical linear regression engine computing 30-day revenue and growth forecasts.
✓ Integrated streaming Google Gemini chat interface allowing users to converse naturally with their raw dataset.

04 // Architectural Decisions & Engineering Trade-Offs

Problem → Decision → Trade-off → Result

DECISION 01 / Client-Side Streaming Ingestion & Local Linear Regression

The Problem:

Small businesses cannot afford enterprise BI suites, but uploading sensitive financial spreadsheets to third-party servers introduces privacy and compliance barriers.

Architecture Decision:

Deployed PapaParse for chunked, in-browser CSV/Excel parsing and executed 30-day linear regression forecasting directly on the client web worker. Data remains strictly on the user's device unless they explicitly trigger AI conversational analysis.

Engineering Trade-Off:

Computations depend on client device CPU capacity.

Measured Result:

Instant dashboard hydration in under 100ms with zero server storage costs and absolute financial data privacy.

DECISION 02 / Streaming Summarization for Google Gemini Context Framing

The Problem:

Large transaction datasets easily exceed context limits or generate high inference latency when sent raw to large language models.

Architecture Decision:

Implemented an automated preprocessing layer that computes aggregated monthly metrics, top revenue categories, and statistical anomalies before injecting a structured JSON digest into the Gemini Flash streaming prompt.

Engineering Trade-Off:

Granular single-line item queries require targeted query filters rather than full raw dumps.

Measured Result:

Real-time conversational insights streamed back in sub-400ms first-token latency with zero context overflow.

05 // Platform Screenshots & Telemetry

Live SaaS Landing & Predictive Architecture (https://new-way-proj.vercel.app) ● LIVE CAPTURE

↳ Live SaaS Landing & Predictive Architecture (https://new-way-proj.vercel.app)

Interactive Financial Analytics Dashboard & Gemini Narrative Insights ● LIVE CAPTURE

↳ Interactive Financial Analytics Dashboard & Gemini Narrative Insights

Drag & Drop Dataset Upload Interface with Instant Schema Detection ● LIVE CAPTURE

↳ Drag & Drop Dataset Upload Interface with Instant Schema Detection

06 // Technologies Deployed

Each technology was selected based on strict performance constraints, team ergonomics, and production reliability.

Next.js 16 React 19 Tailwind CSS v4 Google Gemini Vercel AI SDK Recharts PapaParse TypeScript Zod