Novanda Fadly — Portfolio

01

DompetKu

Personal finance management app

Live

A zero-based budgeting app built with Next.js and Supabase. Manage multiple wallet pockets (operasional, tabungan, kantor), track debts and receivables with wallet integration, and monitor monthly budgets — all in one dashboard.

Designed around a strict architectural rule: all wallet balance changes go through database triggers only — the frontend only inserts or deletes, never updates directly.

▸

Wallet PocketsSeparate buckets for personal, savings, and work expenses with real-time balances

▸

Debt & ReceivablesTrack who owes you (and vice versa) linked directly to wallet movements

▸

Budget TrackingMonthly spend vs budget with category breakdown — kantor expenses excluded from personal budget

▸

Trigger-based ArchitectureBalance integrity enforced at DB level via Supabase Postgres triggers

dompetku.novandafadly.com

Ringkasan Bulan Ini Apr 2026

Operasional

Rp 2.5jt

Tabungan

Rp 1.8jt

Pengeluaran

Rp 1.2jt

Makan siang-Rp 45k

Gaji bulan ini+Rp 8.0jt

Sewa kos-Rp 2.0jt

Tabungan darurat+Rp 1.5jt

02

Reimburse App

Team reimbursement tracker with AI receipt parsing

Internal

An internal web app used by a small team (5 staff) to manage reimbursement submissions. Staff upload receipts and evidence, the app uses Gemini AI to parse them automatically, and generates a clean PDF rekap for approval.

Supports two attachment types: nota (receipts) and evidence rapat (meeting evidence) — each displayed separately in the recap view with proper page-break handling for printing.

▸

AI Receipt ParsingGemini 2.5 Flash extracts amount, vendor, and date from uploaded nota images automatically

▸

Dual Attachment TypesSeparate handling for receipts and meeting evidence with correct labeling in recap

▸

PDF Rekap GeneratorPrint-ready recap with evidence displayed before the summary table, page-break safe

▸

Supabase StorageSecure file storage with Auth — one shared admin account for the whole team

reimburse-app.novandafadly.com

Reimbursement — April 2026

Total Entry

12

Disetujui

8

Total (Rp)

4.2jt

Staff A — TransportRp 85kApproved

Staff B — Rapat PertemuanRp 320kPending

Staff C — ATKRp 45kApproved

Staff D — KonsumsiRp 150kApproved

03

Market Intelligence Agent

AI portfolio research agent with RAG

Live

An AI-powered research agent for IDX (Indonesian Stock Exchange) portfolio analysis. Combines LLM reasoning with real-time financial data and a vector knowledge base to answer complex investment questions about IDX-listed companies.

Built with LangChain/LlamaIndex for agentic orchestration, Supabase pgvector for semantic search over financial documents, and yfinance for live price and fundamental data.

▸

RAG PipelineFinancial reports and analyst notes embedded into pgvector for semantic retrieval

▸

Agentic ToolsAgent can call yfinance, screen stocks by P/BV / ROE / DER, and reason across multiple sources

▸

IDX ScreenerFilters IDX80 + sectoral expansion by custom fundamental criteria

▸

Next.js Chat UIConversational interface for portfolio Q&A with streamed responses

market-intelligence.novandafadly.com

user > Screener IDX80, P/BV < 1.2, ROE > 12%, DER < 1.5

── Calling tool: screen_stocks(filters={p_bv: "<1.2", roe: ">12%", der: "<1.5"}) ──

agent > Found 7 candidates:

BBNI · P/BV 0.95 · ROE 14.2% · DER 0.8

PTBA · P/BV 1.08 · ROE 22.4% · DER 0.4

INDF · P/BV 1.11 · ROE 13.1% · DER 0.9

+ 4 more...

user > Analisis BBNI — layak beli sekarang?

── Retrieving context from pgvector: BBNI Q4 2025 laporan... ──

agent > BBNI saat ini diperdagangkan di bawah book value...

04

PDF Organizer

Client-side PDF manipulation toolkit

Live

A fully client-side PDF tool built with Next.js 15 — no server, no uploads, no privacy risk. Drop in your PDFs, rearrange pages with drag-and-drop, merge multiple files, split, rotate, delete pages, and compress output in three quality presets.

Built with pdf-lib for manipulation and pdfjs-dist for rendering, with a performance-first architecture: lazy page rendering, virtual grid for large documents (200+ pages), and chunked file reading.

▸

Merge & ReorderDrag-and-drop to reorder pages across multiple PDFs before merging into one

▸

Split & CompressExtract page ranges and compress with three presets: screen, ebook, and printer quality

▸

Performance OptimizedVirtual grid + IntersectionObserver lazy rendering — handles 200+ page documents smoothly

▸

100% Client-SideAll processing in the browser — files never leave your device, no backend needed

pdf-organizer.novandafadly.com

PDF Organizer

Merge Split Compress

p.1

■

p.2

□

p.3

□

p.4

□

+

3 files · 12 pages total

↺ Rotate ⬇ Export PDF

05

Real-Time Streaming Pipeline

High-throughput event streaming & OLAP analytics across a private VCN

Live

A high-throughput real-time data platform for transaction event streaming, peaking at ~10,962 transactions/second. Built around Redpanda (Kafka-compatible, JVM/ZooKeeper-free broker) feeding a ClickHouse OLAP store, with sub-second analytical queries surfaced through a live Metabase dashboard.

Deployed on OCI across two VMs communicating over a private VCN subnet: a DB VM running ClickHouse, Redpanda, and Nginx, and an app VM running Metabase for dashboarding. Next up: a GCP BigQuery cold-storage warehouse bridged via a Python ETL batch service, for unified hot + historical querying.

Note: this is an architecture case study — the live dashboard runs on simulated transaction data, not real payment traffic.

▸

Redpanda IngestionKafka-compatible streaming broker handling peak throughput of ~10,962 TPS with topic partitioning & retention strategy

▸

ClickHouse OLAP StoreRaw transaction table, hourly aggregations via SimpleAggregateFunction, and a dedicated fraud-detection log for sub-second analytical queries

▸

Dead Letter QueueIsolates failed events into a DLQ table so pipeline failures never block the main streaming flow

▸

BigQuery Warehouse (in progress)Planned Python ETL service to sync ClickHouse hot storage to BigQuery cold storage for unified historical querying

analytics.novandafadly.com/dashboard/real-time-payment

stream > Redpanda producer → ClickHouse rtp_analytics · auto-refresh 10s

── SELECT count(), sum(amount) FROM user_transactions WHERE created_at >= NOW() - INTERVAL 10 SECOND ──

throughput > Peak: 10,962 TPS

status: success · failed · pending — 24h breakdown

fraud_alerts: high_value_fraud_alerts — real-time rule engine

dlq_raw_transactions: 0 unresolved failures

etl > Planned: batch sync ClickHouse → BigQuery (cold storage)

── google-cloud-bigquery SDK · not yet connected ──

status > Hot storage live on OCI · BigQuery integration upcoming

Novanda
Fadly.

Analyst by day,
builder by night.

Things I've built.

Skills & tools.

Credentials.

Let's connect.

NovandaFadly.

Analyst by day,builder by night.

Things I've built.

Skills & tools.

Credentials.

Let's connect.

Novanda
Fadly.

Analyst by day,
builder by night.