Forward deployed AI engineerOpen to remote

LaskarKsatria Satika

I take AI systems from a client's brief to something their customers use.

LLM engineer and senior Shopify engineer. Leading eight engineers at Antikode, based in Jakarta.

Fig. 01 — Project map13 stars
Every star is a project. Bigger means further along, from published model up to live in production.
Jakarta, ID7 yrs shipping softwareLead, AntikodeScroll

Clients and milestones: Wardah, Bodypack, Makeover, Sepatu Compass, Qwen3.5-2B fine-tune, 0 downtime.

02About

Between the customer and the code

I've spent about seven years shipping software for Indonesian consumer brands. Most of it starts as a client brief and ends as something their customers touch every day — storefronts, discount engines, skin analysis, and now agents and fine-tuned models.

I build agents harness-first: the model handles language and judgment, typed code handles tools, memory and checks.

Before software I spent six years at Krakatau Engineering and Pertamina EP, so I know how enterprise customers work.

018engineers I lead at Antikode
020downtime during a national flash sale
034 GBRAM budget for my fine-tuned model in a live game
R1Forward deployed AIWardah skin analysis, live for a national beauty brand.
R2LLM engineeringFine-tuned Qwen models, evaluated against real baselines.
R3Shopify engineeringHeadless Hydrogen storefronts and custom discount backends.
04Index of work

Every star, itemised

01Makeovermakeoverforall.comIn productionProduction work for Makeover, an Indonesian cosmetics brand.
02Wardah skin analysisAI agents for beauty retailIn productionSelfie skin analysis live in Wardah's Skin Dictionary. Python/FastAPI with AutoGen agents.
03Bodypack discountsShopify engineering · custom backendIn productionFlash sales, automatic Buy X Get Y and bulk discounts beyond Shopify's Growth plan. Zero downtime at peak.
04Compass storefrontHeadless HydrogenIn productionHeadless Hydrogen storefront for one of Indonesia's leading sneaker brands.
05AI GladiatorSolo build, open betaLive betaTrivia battle game on my own Qwen3.5-2B fine-tune. Train it, then send it to fight.06Orbit LabAgent harness engineeringDemoTwo business agents in a sandbox: bank reconciliation and field-note work orders, with tools, memory, loop and per-step checks exposed.
07Altair agentSite-audit agentInternalSite-audit agent: the LLM routes and narrates, typed code crawls and scores.
08Floor 7thMulti-agent demoDemoFour agents research, draft and edit an article while you watch every tool call live.
09Pyxis Stock AnalystAI stock analysis · milky-way.laskarks.comDemoUS stock analysis with AI: signals, bull and bear cases, and a chat with memory.
10RerankerLoRA fine-tuneModelPicks a product's category from a retrieved shortlist: 66% accuracy vs 34% for the same model zero-shot.
11AlcyoneGPT from scratchModel30.5M-parameter GPT pretrained from scratch on a single T4.
12GuardrailToxicity classifierModelIndonesian/English toxicity filter, ONNX INT8 at ~75 MB, runs on CPU. Macro-F1 0.825 (ID).
13Revenue estimatorRegression modelModelEstimates revenue per employee from a company description. Beats TF-IDF and linear baselines.
05Stack

What I build with

Picked for the problem in front of me, grouped by where I use them.

AI & agents01
MastraAgent framework
AutoGenMulti-agent
LangGraph.jsAgent graphs
MCPTool protocol
RAGRetrieval
EvalsPer-step checks
Models02
QwenFine-tuned base
LoRA / QLoRAFine-tuning
UnslothTraining
HF TransformersTraining, inference
ONNX INT8CPU inference
OllamaLocal serving
Backend03
Node.jsAPIs, workers
PythonML pipelines
FastAPIPython APIs
HonoEdge APIs
PostgreSQLDatabase
MongoDBDatabase
Frontend04
Next.jsApp Router
ReactUI
TypeScriptEnd to end
PWAInstallable apps
Commerce05
ShopifyStorefronts
HydrogenHeadless
Storefront APICart, checkout
Custom discountsNode.js + MongoDB
06WritingAll posts
07Contact

Hiring a forward deployed engineer?

Open to FDE, AI engineer and LLM engineer roles. Freelance briefs welcome too.

© 2026 Laskar Ksatria SatikaJakarta, ID · Remote OK