Eggi Satria EgiStr

Computer Vision · LLM & RAG · Applied AI

I build AI that works outside the lab — from CCTV streams on a packing-house conveyor to RAG assistants and agent tooling. I care about the unglamorous half of the job: calibration, data contracts, failure gates, deployment, and updates that roll back when they go wrong.

🔭 What I'm doing now

🎯 AI Engineer (Computer Vision) — real-time detection, counting, sizing & grading from CCTV
🧠 LLM Apps & RAG — retrieval pipelines, evaluation harnesses, agent skills
🧪 Focus — making CV/LLM systems measurable and reproducible
🎓 Sains Data @ ITERA — GPA 3.75/4.0

⚡ Quick facts

📍 Indonesia (UTC+7)
🗣️ Indonesian · English
🔧 Default stack: Python, PyTorch, FastAPI, Docker
🧩 I like systems: pipelines, threading, OTA, observability
📫 Always open to talk applied AI

🚀 Featured Work

🥑 GuavaCam — Industrial CV System

Real-time guava detection · counting · sizing · grading from CCTV on a 3-lane packing-house conveyor.

YOLO segmentation + IoU tracking
Checkerboard calibration → ray-plane pixel-to-mm
XGBoost weight estimation (A3 core-10 features)
Multi-threaded runtime, async DB writer w/ dead-letter, OTA updates with health-gate + auto-rollback

Python PyTorch YOLO OpenCV XGBoost MySQL

🪖 SmartCampus Vision — Traffic Analytics

Real-time helmet-violation detection & tracking on an event-driven pipeline.

YOLOv8 + ByteTrack, end-to-end ML workflow
Kafka → Spark Streaming event pipeline
Star-schema data mart in PostgreSQL + Grafana dashboards

PyTorch YOLOv8 ByteTrack Kafka Spark PostgreSQL

⚖️ CourtSight — AI Legal Intelligence

International Hackathon Finalist (AngelHack PAN-SEA AI Developer Challenge).

RAG over legal documents w/ pgvector + hybrid search
LangGraph orchestration for legal reasoning
Full production deployment on GCP

LangGraph RAG pgvector FastAPI GCP

🆔 Indodoc Vision — Document Extraction

Automated Family Card (KK) field extraction.

Hybrid architecture: YOLOv8 locates field regions, VLM reads text — avoids VLM hallucination on dense tables
Clean structured JSON output, served via REST API

YOLOv8 VLM FastAPI OCR

🌴 OilPalmVision.ai — Plantation Census

SaaS platform for automated oil-palm tree census.

Handles 2GB+ orthomosaic files via auto dataset tiling
Pixel → WGS84 GPS coordinate conversion
FastAPI + Redis Queue for async detection jobs

YOLOv8 OpenCV FastAPI Redis

🛠️ Agent Skills & Tooling

Reusable skills/tools I built and open-sourced.

keyless-research-mcp — academic research MCP, zero API keys
android-agent-skills — QA Android phones over ADB as text
motion-reel — narrated motion-graphics video from HTML
cariak — deep research agent, sourced claims

Python MCP TypeScript MIT

🧰 Tech Stack

AI / Machine Learning

Backend / Data / DevOps

📊 GitHub Stats

🔬 Currently Exploring

Metric vision — tightening calibration so pixel→millimetre error stays predictable across shifts
Agent reliability — measuring when a retrieval pipeline is actually grounded, not just fluent
Edge inference — squeezing real-time throughput out of hardware that was never meant for it

_{Updated whenever the work moves. Longer notes land on LinkedIn.}

🏆 Highlights

💬 Let's build something that actually ships

_{Updated 16 Oct 2026, 10:20 WIB}

Navigation Menu