# Candra Rudyatmoko AI Engineer | Full-Stack Developer | LLM Application Developer ## Professional Summary Software engineer with a background in full-stack product delivery and a strong recent focus on AI applications. Hands-on experience building LLM-powered developer tools, multi-provider AI platforms, OCR pipelines, and NLP experimentation workflows for Indonesian-language use cases, backed by prior delivery experience across mobile applications, GIS platforms, dashboards, CMS, public-sector systems, and enterprise web applications. Core strengths include designing agentic systems that can interact with files, shell commands, databases, and project context; integrating multiple AI providers into a single product experience; and combining AI models with classical NLP, OCR, and rule-based processing to solve real operational problems. This AI work is reinforced by practical engineering experience delivering production-oriented systems for government and organizational workflows. ## Core Competencies - AI agents and tool calling - Multi-provider LLM integration - Developer tooling and workspace automation - OCR and document intelligence - Indonesian NLP experimentation - Embedding, clustering, and topic analysis - Local LLM experimentation and GGUF runtime - Local model serving with llama.cpp / llama-cpp-python - API and backend development - Python, JavaScript, and TypeScript - FastAPI, Flask, SQLite, and automation workflows - Flutter mobile application development - GIS and geospatial web platforms - Public-sector and enterprise application delivery ## Selected AI Experience ### Siberflow Repository: `https://github.com/candrapwr/siberflow` Built an AI platform that supports multi-provider model access, streaming tool calling, file sandboxing, database access, multi-session persistence, and task checklist management through CLI and VS Code interfaces. Key contributions reflected in the repository: - designed an agent loop architecture for AI-driven workflows - integrated multiple model providers including DeepSeek, Gemini, OpenAI, Grok, and Qwen - enabled AI interaction with files, shell commands, and databases - implemented cross-interface session persistence between CLI and VS Code extension - added context optimization and auto-continue behavior for more robust user experience Technology stack: - TypeScript - Node.js - npm workspaces - CLI tooling - VS Code extension architecture ### IdSiberAi-CLI Repository: `https://github.com/candrapwr/IdSiberAi-CLI` Developed a command-line AI assistant that combines multiple LLM providers with local system management capabilities for technical productivity and automation use cases. Key contributions reflected in the repository: - built a CLI assistant for practical computing and automation tasks - integrated multiple AI providers with provider switching and fallback behavior - connected AI workflows to file operations, storage tasks, code analysis, and database tools - included streaming response handling and operational logging - provided both CLI and web-based interaction modes Technology stack: - JavaScript - Node.js - Multi-provider LLM integration - S3 integration - MySQL, PostgreSQL, and SQLite tooling ### IdSiberCoder-VSCode Repository: `https://github.com/candrapwr/IdSiberCoder-VSCode` Built a VS Code extension that brings an AI developer agent directly into the editor workspace, allowing the model to inspect files, search context, execute commands, and edit project files through tool calling. Key contributions reflected in the repository: - implemented a workspace-aware AI coding assistant inside VS Code - built configurable tool management through an interactive settings panel - supported multiple AI providers - enabled parallel tool execution for selected actions - added safeguards such as iteration limits, timeouts, and output sanitization Technology stack: - JavaScript - VS Code Extension API - Webview integration - Tool registry and schema-driven actions ### information_extraction Repository: `https://github.com/candrapwr/information_extraction` Created an OCR pipeline for extracting structured information from Indonesian identity cards and international passports using OpenCV, Tesseract, EasyOCR, and optional LLM-assisted extraction. Key contributions reflected in the repository: - built preprocessing steps to improve OCR stability - supported multiple OCR engines under a unified workflow - added heuristics to handle noisy OCR outputs - exposed the workflow through both CLI and Flask API - combined OCR, parsing logic, and optional LLM integration into one practical pipeline Technology stack: - Python - OpenCV - Tesseract - EasyOCR - Flask ### media_analytics Project location: `/Volumes/Workspace/python_app/media_analytics` Developed and explored an AI/NLP experimentation workspace focused on Indonesian-language media analysis, including data scraping, text preprocessing, embedding generation, clustering, keyword extraction, visualization, chatbot experiments, and local LLM inference. Key contributions reflected in the project files: - built a scraping and SQLite-based dataset pipeline for news content - applied Indonesian text preprocessing with NLTK and Sastrawi - used IndoBERT for embedding and topic-analysis experiments - combined TF-IDF, KMeans, and KeyBERT for clustering and keyword extraction - generated text and visual reports for trend analysis - experimented with simple chatbot workflows using IndoBERT QA and mT5 - tested local inference workflows using Hugging Face models and GGUF runtime - wrapped local GGUF models behind OpenAI-style chat endpoints for lightweight integration experiments - explored lightweight PyTorch-based generative text training Technology stack: - Python - SQLite - BeautifulSoup - NLTK - Sastrawi - scikit-learn - Transformers - IndoBERT - KeyBERT - FastAPI - PyTorch - GGUF runtime ### Local GGUF and llama.cpp Experimentation Project locations: - `/Volumes/Workspace/python_app/gguf` - `/Volumes/Workspace/python_app/media_analytics/gguf` In addition to cloud-model integration, there is clear experimentation with local model execution using GGUF models and `llama.cpp` / `llama-cpp-python`. The visible files show hands-on work with local chat inference, API wrapping, and multiple quantized models across general LLM, coding, OCR, and vision-language use cases. What is visible from the workspace: - local GGUF model collection including Gemma, Qwen, DeepSeek-R1, GLM, LLaVA, Qwen-VL, coder, and OCR-oriented models - `llama-cpp-python` scripts for direct local inference - FastAPI wrapper that exposes a local model through an OpenAI-style `/v1/chat/completions` endpoint - interactive local chat scripts with prompt construction and optional conversation history - notes for building and running `llama.cpp` with Metal acceleration on Apple Silicon - experimentation across text, coding, OCR, and multimodal/vision-capable model formats What this adds to the profile: - practical familiarity with local LLM deployment, not only hosted APIs - understanding of quantized model tradeoffs, context sizing, and lightweight serving patterns - experience bridging local inference into developer-friendly interfaces and API-compatible workflows - stronger credibility for privacy-sensitive, offline-capable, or cost-aware AI deployment scenarios ## Supporting Engineering Projects ### Video-Generator-Service Repository: `https://github.com/candrapwr/Video-Generator-Service` Built a media automation service for generating dynamic slideshow videos from images, videos, and audio through FastAPI and a web interface. This project demonstrates practical backend engineering for content workflows that can complement AI-driven media generation systems. ### Audio-Mixer-Service Repository: `https://github.com/candrapwr/Audio-Mixer-Service` Built an automated audio mixing service with dynamic ducking, silence detection, timing control, trimming, API access, and web UI. This project demonstrates engineering experience in workflow automation for creative and media processing use cases. ## Broader Software Engineering Experience In addition to AI-focused work, the portfolio PDF shows delivery experience across more than twenty software projects, especially in government and organizational environments. These projects strengthen the overall profile by demonstrating that the AI work sits on top of substantial application-development experience rather than in isolation. Key areas visible in the portfolio PDF: - Flutter mobile applications for Kementerian Perdagangan, including E-Kemendag, G20 TIIWG Mobile, and Hero Mobile - web-based internal systems and portals using PHP, CodeIgniter, Laravel, MySQL, PostgreSQL, SQL Server, MSSQL Server, and OracleDB - GIS and map-based systems using LeafletJS and GeoServer, including stock maps, interactive maps, geoportals, and Covid-19 monitoring - dashboards, monitoring systems, recruitment portals, project management systems, attendance systems, licensing systems, and tax-payment systems - CMS and content-management workflows for institutional portals - version control practices using Git across multiple projects Representative non-AI project themes from the PDF: - mobile service applications for internal government workflows - GIS platforms for public information and spatial monitoring - operational systems for recruitment, testing, licensing, attendance, and financial monitoring - enterprise and institutional dashboards for decision support and reporting ### Selected Non-AI Delivery Experience The non-AI portfolio is strongest when presented as several consistent delivery tracks rather than a long undifferentiated project list. #### 1. Mobile Applications for Government Workflows Projects visible in the PDF include: - E-Kemendag mobile application for Kementerian Perdagangan - G20 TIIWG mobile application - Hero Mobile help-center application What this shows: - experience delivering Flutter-based mobile applications - work on internal service flows, authentication-oriented environments, and structured feature delivery - familiarity with production-oriented mobile product development for institutional users #### 2. GIS and Geospatial Information Systems Projects visible in the PDF include: - Monitoring Pohon KPHP DLHK Provinsi DIY - Geoportal Pemerintah Kota Yogyakarta - Peta Interaktif Pemerintah Kota Yogyakarta - Peta Stok Provinsi SISP - GIS Monitoring Sebaran Covid-19 Kota Yogyakarta - Portal Manajemen Peta Gowes Sepeda Kota Yogyakarta What this shows: - strong experience building map-centric platforms and spatial data interfaces - use of LeafletJS, GeoServer, PostgreSQL, and related GIS-oriented web stacks - ability to design systems that combine data management, visualization, and operational monitoring #### 3. Operational and Administrative Systems Projects visible in the PDF include: - Sistem Computer Based Test Kementerian Perdagangan - Portal dan CMS Rekrutmen ASN Kementerian Perdagangan - Sistem Perizinan Elektronik DINPMPTSP Kabupaten Purworejo - Sistem Absensi Pegawai Elektronik DINPMPTSP Kabupaten Purworejo - Sistem POS Pembayaran PBB Kota Samarinda - Sistem Monitoring Dana Kelurahan Kota Yogyakarta What this shows: - experience building business-critical workflow systems used for recruitment, testing, licensing, attendance, payments, and budget monitoring - familiarity with form-heavy platforms, administrative dashboards, role-based workflows, and institutional reporting requirements - practical delivery background in systems where reliability and process clarity matter #### 4. Dashboards, Monitoring, and Reporting Platforms Projects visible in the PDF include: - Modul Monitoring Data Pasar SISP - Dashboard Monitoring Kunjungan Kota Yogyakarta - Portal Manajemen Kliping Kota Yogyakarta - project monitoring and management systems for institutional use What this shows: - experience with analytics-style internal tools and monitoring dashboards - ability to structure backend data flows into usable operational views - familiarity with reporting-oriented applications for decision support #### 5. CMS and Content Platforms Projects visible in the PDF include: - CMS Portal Reborn Kominfo RI - CMS Rekrutmen ASN Kementerian Perdagangan What this shows: - experience building and maintaining content-management workflows - work on systems that require editor usability, structured publishing, and administrative control - broader product thinking beyond pure backend implementation #### 6. Enterprise and Institutional Delivery Context Across the PDF, the client and domain patterns include ministries, local governments, and institutional platforms. What this adds to the profile: - experience delivering software in formal organizational environments - ability to work across varied stacks including PHP, Laravel, CodeIgniter, Flutter, Vue, PostgreSQL, MySQL, SQL Server, MSSQL Server, OracleDB, LeafletJS, and GeoServer - evidence of adapting engineering work to public-sector and enterprise-style requirements What this adds to the positioning: - evidence of shipping software beyond experiments and prototypes - experience working on business-critical and institution-facing systems - stronger credibility for moving AI solutions into real production contexts - stronger AI systems credibility through both cloud-provider and local-model workflows - broader range across mobile, backend, database, GIS, and web application development ## Positioning The strongest professional positioning based on the visible body of work is: - AI Engineer - Full-Stack Developer with AI specialization - LLM Application Developer - AI Tooling Engineer - Software Engineer for AI and Workflow Automation ## Short Bio AI engineer and full-stack developer focused on LLM applications, AI agents, OCR pipelines, Indonesian NLP experimentation, local model workflows, and developer tooling. Combines recent AI product development with broader experience delivering mobile apps, GIS systems, dashboards, and public-sector web platforms.