M Sechan Alfarisi, Full-Stack Mobile Software Engineer and Applied AI Engineer

The problem

Most reading trackers only log page counts and give readers no reason to come back. Progress across chapters is hard to follow, and reflections end up scattered in separate note apps with no community around them.

What I built

1. 01Wrote 286 of 326 non-merge commits across Flutter (GetX, Clean Architecture: DataSource, Repository, UseCase, Controller, UI) and Supabase.
2. 02Built the reading activity tracker (pages, chapters, percentage, daily reflection) on an atomic Postgres function, record_reading_activity_atomic, so two updates arriving together cannot reset a book's progress.
3. 03Built the Book Reflection Studio: an editorial layout with Newsreader and Inter type, floating section navigation and an embedded 16:9 YouTube player.
4. 04Built a tarot-themed badge system with celebration animations and a three-badge showcase on reader profiles.
5. 05Fixed a Postgres 22P02 crash in the genre filter by rewriting the JSONB query with the containment operator.
6. 06Tightened Supabase row-level security and reduced release size with WebP assets and ProGuard shrinking.

Stack

The problem

Small businesses in cities like Sukabumi need digital promotion but cannot risk paying for ads up front. Marketiv offers two modes: pay-per-view campaigns with no chat, and fixed-price rate cards with escrow and chat.

What I built

1. 01Lead the web platform in Next.js 16, React 19, TypeScript and Tailwind CSS v4. Authored 231 of 398 non-merge commits in the main repository.
2. 02Separated the MSME and creator dashboards at the layout level so they share no client state, checked by an automated data-contract script.
3. 03Co-built the Appwrite backend with the CEO: Cloud Functions for Midtrans payments, escrow, creator withdrawals with a balance guard, and an AI campaign brief generator on Vertex AI.
4. 04Added idempotency keys to checkout so a double tap or a retry on a weak connection cannot create a second payment.
5. 05Matched Zod schemas to Appwrite column sizes so long forms show field-level errors before submitting instead of a generic 400.
6. 06Moved the admin dashboard into its own Next.js app with fail-closed role checks (admin, staff).

Stack

The problem

Shoppers in physical stores often only learn their total at the cashier. Typing every item into a calculator while holding a basket is slow, and mobile signal inside stores is often too weak to upload audio.

What I built

1. 01Split the work between device and cloud: speech recognition runs on the phone with expo-speech-recognition, and only the transcript goes to the LLM.
2. 02Wrote few-shot prompts for Groq (Llama 3.1 8B) and Gemini that parse local price slang such as "goceng" and "ceban" and tell unit price from total price, returning typed JSON.
3. 03Stored carts and history in Expo SQLite so the app works with no connection.
4. 04Added budget warnings with color changes and haptic feedback, plus automatic failover to a backup API key on rate limits.
5. 05Wrote the paper and presented it at SENTIMETER 2026.

Stack

The problem

The clinic ran patient registration, queue, doctor records, BPJS and general billing, the cash book and reports on separate Google Sheets and Excel files. Years of legacy data sat in CSV files of uneven quality.

What I built

1. 01Designed and built the whole system alone: architecture, database schema, UI and deployment.
2. 02Wrote a spec for each feature before building it, plus four architecture decision records.
3. 03Migrated 7,493 legacy rows into PostgreSQL, handling mislabeled columns and medical record numbers of mixed length.
4. 04Made the reconciliation script deterministic, so re-running it on production plans zero changes. Possible duplicate patients are left for the clinic to decide instead of being merged automatically.
5. 05Protected patient data with 17 versioned SQL migrations, row-level security and two roles (owner, doctor-admin).
6. 06Built the dashboard, Excel export, Puskesmas reports, health-program registers, therapy packages and queue cancellation.

Stack