00 BACKEND · AI · FULL STACK

FULL STACK DEVELOPER

BASEDTANGERANG · WIB

DEPTHARCHITECTURE · AI

SINCE2023

MODECONTRACT / FREELANCE

I build the boring-but-critical stuff: backend services, RAG pipelines, auth layers. The parts that just need to work. Informatics Engineering at Politeknik Multimedia Nusantara, 3.94 GPA, recognized by Samsung Innovation Campus.

SCROLL

PYTHON/FASTAPI/GO/POSTGRESQL/PGVECTOR/REDIS/DOCKER/NEXT.JS/NATS JETSTREAM/CLEAN ARCHITECTURE/JWT & OAUTH2/

APPROACH

I build for
later, not
for demo day.

Easy when it's just you writing to it. Gets messy once two teams need the same data to line up, or someone who's never touched a foreign key has to make sense of your schema.

So I keep it boring on purpose: real schemas over caches, real RBAC over shared logins, idempotent jobs over retry buttons.

Backend-first, but I ship the whole thing, schema to handover, so nothing quietly breaks in between.

$ make report --since=2023 BUILD PASSED

[ OK ]years building software3

[ OK ]systems shipped to production12

[ OK ]tables in my biggest schema yet37

[ OK ]cumulative gpa3.94/4.0

[ OK ]peak concurrent users, load-tested100K+

DIFF ~/asset-systems commit 4a9e21c

-flat tables, no audit trail on transfers

-asset returns tracked across work units by hand

+37-table schema, one source of truth

+returns as a state machine, audit log built in

SELECTED WORK

TWO SYSTEMS,
END TO END.

CASE 01 · GOVERNMENT & ENTERPRISE · 2023-PRESENT

State Asset & Inventory Backend Systems

ROLE · DATA MODEL · BACKEND · TRAINING

FASTAPI POSTGRESQL SQLALCHEMY DOCKER

01 / PROBLEM

Asset records lived in flat tables with zero audit trail. Everyone requested transfers their own way, so nothing ever matched up.

BOTTLENECK · NO AUDIT TRAIL ON RETURNS

02 / ARCHITECTURE DECISION

Schema first, API second. One source of truth instead of a dozen exports. FastAPI handles auth and the whole asset lifecycle, no more manual reconciling.

37 TABLES · 1 SCHEMA

03 / IMPLEMENTATION

PostgreSQL mapped to how assets actually move: in, held, transferred, returned. SQLAlchemy handles the relations, Docker keeps it reproducible.

NEW SCHEMA · LEGACY MIGRATED

04 / OUTCOME

One schema instead of a workaround per unit. New features ship without breaking the trail, and I train the staff on it myself.

LEGACY → REBUILT

EDGE

WORK UNITSasset requests

PAPER FORMSno schema

LEGACY REGISTRYflat tables, no trail

STATE INVENTORYlegacy · read-only

GATEWAY

ASSET SERVICE APIFastAPI · OpenAPI

AUTH · JWT

RBAC · ROLE-SCOPED

AUDIT · APPEND-ONLY

DOMAIN

ASSET REGISTRYhierarchy · custody

ASSET TRANSFERunit-to-unit

RETURN WORKFLOWstate machine

STAFF PORTALREST · web

TRAINING DOCSguides · specs

DATA

POSTGRESQL37-table relational schema37 TABLES

ASYNC

REDIS CACHEsession · lookups

BACKGROUND JOBSreports · exports

OUTCOME PACKET RATE FOLLOWS SCROLL VELOCITY →

CASE 02 · HIGH-CONCURRENCY SYSTEMS · 2025

CONCERT TICKET QUEUE SYSTEM

A queue for concert ticket sales that stays fair under 100k+ concurrent users and never sells the same seat twice.

GO REDIS NATS JETSTREAM POSTGRESQL

STEP 01 / 05 · SCROLL RUNS HORIZONTAL →

02 / CONSTRAINT

The queue has one job: nobody double-checks out.

Every stage had a load-test budget before I wrote a single handler. Too slow under k6 at 100k users, it got moved off the hot path.

JOIN QUEUE8ms

VALIDATE14ms

LOCK SEAT19ms

CHECKOUT62ms

HEADROOM300ms

03 / PIPELINE

One direction, one lock, no double-checkouts.

01REQUEST INTAKENATS JetStream · dedupe

02VIRTUAL QUEUERedis sorted set

03POSITION CHECKpure fns · unit-tested

04SEAT LOCKoptimistic lock · fails closed

05CHECKOUT WORKERidempotent ids · replay

Each stage talks through NATS, not a shared function, so I can restart or swap one without touching the rest.

04 / GUARDRAILS

A queue system is judged on how it says no.

KILL SWITCHOne flag, checked before every checkout, honoured before any hold.

SEAT CAPSPer-event and per-user ceilings, enforced server-side, not in the client.

OVERSELL CIRCUITAny mismatch between seats held and seats sold trips the breaker.

REPLAY LOGEvery checkout reproducible offline from its exact inputs.

05 / OUTCOME

18MSP99 QUEUE-JOIN, K6 @ 100K VUS

100K+CONCURRENT USERS, LOAD-TESTED

0DUPLICATE CHECKOUTS IN LOAD TEST

Its best feature does nothing: it just holds you in line instead of letting two people fight over a seat. That's the whole point.

ALSO SHIPPED

01AUTH & IDENTITY

One SSO, every internal app

POST /auth/login → issue JWT (rolling)

RBAC · scoped roles per tenant

AES-256 · encrypted responses

rate limit · 100 req/min per IP

Multi-tenant SSO with rolling JWTs, RBAC, and AES-256 encrypted responses. One identity layer instead of everyone rolling their own.

MULTI-TENANTROLLING JWTAES-256

02AI / RAG

A RAG system with no config step

connect → any PostgreSQL schema

sync → real-time, no manual mapping

query → hybrid SQL + vector search

answer, natural language, sourced

Points at any PostgreSQL database, reads the schema itself, stays in sync, answers questions in plain English.

PGVECTORHYBRID SEARCHANY SCHEMA

03PAYMENTS

QRIS checkout that can't double-charge

create charge → poll Xendit status

→ on paid, dedupe by charge id

→ upload Cloudinary, async

→ notify SMTP receipt, auto-retry

QRIS payments for photobooth kiosks. Polls for status, dedupes charges, emails a receipt automatically.

XENDITQRIS0 DOUBLE-CHARGES

04TRANSACTIONS

Stock moves that can't half-happen

PO → stock transfer → item request

atomic writes · no partial commits

audit trail · every movement logged

rollback · one failed row, one retry

Warehouse system where POs, transfers, and audit trails share one atomic write path. Nothing gets left half-moved.

ATOMIC TXNSFULL AUDIT TRAIL

EXPERIENCE

MAY 2026 - PRESENT

Full Stack Programmer SEKRETARIAT JENDERAL DPR RI · JAKARTA

Build and maintain applications for the State-Owned Assets Administration Division, including SIRANGGA V2, a redesigned asset inventory & return system. Train work units across the Secretariat General on rollout.

FEB 2025 - JAN 2026

Backend Developer, Intern ALFAGIFT (PT GLOBAL LOYALTY INDONESIA) · TANGERANG

Shipped user-facing web features and modular, reusable components across devices. Monitored performance and traffic to catch usability issues before they became tickets.

2023 - 2024

Developer, Freelance PT SAMANASOFT INOVASI PERSADA · TANGERANG REGENCY

Three engagements: an AI chatbot that streamlined hospital appointment booking, a machine learning model recommending medication dosages, and a round of feature upgrades to an existing program.

ACHIEVEMENTS & CERTIFICATIONS

CERTIFICATIONOCT 2024

International Labor Standards Certification

International Training Centre

Labor laws, business compliance and legal frameworks for workplace regulations.

CERTIFICATIONSEP 2024

Samsung Innovation Campus

Samsung Electronics

Technical innovation, problem-solving methodology and innovation management.

Chairman and Show Director role at Inaugurasi 2024

AWARD · CHAIRMANAUG 2024

Chairman & Show Director, Inaugurasi 2024

Politeknik Multimedia Nusantara

Led creative vision and operations for the university's new-student inauguration.

CERTIFICATIONAPR 2024

Data Science Professional Certification

DQLab

Python, SQL, data visualization and statistical analysis for real-world data.

AWARD · 1ST RUNNER-UPOCT 2023

National Podcast Competition

PELITA

National-level recognition for content creation and communication skills.

CERTIFICATIONJUN 2023

Frappe Framework Bootcamp

Anteraja Logistics

Framework training for logistics platforms and software engineering practice.

$ find ./certificates -name "*nbp-2023*"

no matches found

participation only, no cert issued

COMPETITION2023

National Business Plan 2023

Team SquidWork

Competed in a national business-plan competition as part of team SquidWork. No certificate on file, no rank claimed.

CAPABILITIES

HOVER A CAPABILITY FOR DETAIL

LANGUAGES

Python Go TypeScript SQL C#

FRAMEWORKS

FastAPI Laravel Gin Next.js SQLAlchemy

DATA

PostgreSQL pgvector Redis MongoDB MySQL

AI & INTEGRATIONS

RAG LLM Integration NATS JetStream Webhooks

INFRA & DEVOPS

Docker GitHub Actions k6 Linux

PRACTICE

Clean Architecture Multi-tenancy RBAC JWT & OAuth2 Documentation

NOTES

ASK

READY

Ask about the work above.

A small assistant grounded in what's on this page, roles, projects, stack. Off-topic questions get pointed to email instead of a guess.

$ ./ask-gratia --interactive LLAMA 3.3 · GROQ

AI Hey, ask me anything about Gratia's experience, stack, or projects, SIRANGGA V2, ATABOT, the ticket queue system, any of it.

DIFF ~/engagement working-agreement.md

-we'll figure out the data model as we go

+constraints written down, then the schema, then the code

CONTACT

OPEN FOR FREELANCE BACKEND & AI WORK.

COMMITTED · 9a5ea87 → main

Quickest way in: tell me what you're stuck on. I'll tell you how I'd model it, and what I wouldn't build.

EMAIL gratiamanullang03@gmail.com GITHUB /GratiaManullang03 LINKEDIN /in/felix-gratia-mangatur-manullang

Download CV ↓