A selection of projects where I've paired robust LLM engineering with cinematic product experiences. Each one designed to scale and deliver real user value.
An AI-Powered Presentation Generation Platform
AI-powered presentation assistant that transforms ideas into polished slides with intelligent layout suggestions, content generation, and real-time collaboration features.
Key Insight: Combining vision models with layout generation algorithms enabled contextually relevant design recommendations that reduced presentation creation time by 75%.
A Glass-Box RAG Framework for Pipeline Debugging and Visualization
A comprehensive RAG framework that exposes every step of the retrieval-augmented generation pipeline—from document ingestion and query embedding to vector retrieval and LLM generation. Built with Python 3.11, FastAPI, and LangChain (LCEL), RAGScope provides real-time visualization of retrieved chunks with similarity scores, latency breakdown across stages, token usage tracking, and cost analysis. Features multi-config support for comparing chunking strategies (Recursive vs Fixed), testing different embedding models, and experimenting with chunk size, overlap, and retrieval parameters to optimize RAG performance.
Key Insight: Building glass-box visibility into RAG pipelines revealed that 40% of retrieval failures stemmed from suboptimal chunking strategies. Enabling real-time comparison of configurations reduced pipeline debugging time by 85% and cut API costs through intelligent caching and configuration-based optimization.
An Intelligent LinkedIn Recruitment Automation System
AI-powered recruitment agent that autonomously navigates LinkedIn to discover hiring opportunities and evaluate candidate-job fit. Uses undetected Selenium for authenticated browsing, Gemini API for multi-step reasoning to analyze job posts, extract decision-maker contact information, evaluate candidate profile alignment, and generate personalized outreach emails. Features intelligent pagination, rate limiting to avoid detection, structured data extraction from hiring posts, and semantic matching algorithms to score candidate compatibility.
Key Insight: Implementing multi-agent orchestration with specialized sub-agents for scraping, analysis, and email generation—combined with intelligent retry logic and session management—achieved 92% success rate in extracting hiring leads while maintaining stealth operations and processing 200+ posts per session without triggering LinkedIn's anti-bot measures.
A Vision-Based Nutritional Analysis Engine
iOS app that uses multi-step LLM reasoning to analyze food images and provide accurate calorie tracking. Features advanced image processing, compression optimization, and integration with comprehensive nutrition databases for precise dietary insights.
Key Insight: Implementing multi-step LLM workflows with image compression before API calls reduced processing costs by 60% while maintaining 95% accuracy in food recognition and calorie estimation.
An AI-Powered Relational Intelligence Platform
AI-powered relationship app that acts as an intelligent mediator to generate personalized questions for couples. Features swipeable flashcards for group activities, contextual conversation memory, and adaptive questioning based on relationship duration and topic preferences.
Key Insight: Implementing memory-augmented RAG with intelligent caching reduced response times by 75% while maintaining contextual conversation flow across multiple sessions, creating more natural relationship-building experiences.
A Real-Time Cinematic Conversation Renderer
iOS app that creates cinematic texting story videos like WhatsApp chat simulations. Features dual-side conversation rendering, real-time video generation, and advanced memory management for smooth storytelling experiences.
Key Insight: Implementing parallel rendering pipelines with race condition handling enabled simultaneous dual-character animation while maintaining 60fps performance and optimal memory usage on mobile devices.
A Computer Vision Plant Health Diagnostic System
Mobile app that uses computer vision to identify plants, detect diseases, and provide personalized care recommendations based on environmental conditions and user location.
Key Insight: Fine-tuning a lightweight vision model with a custom plant disease dataset achieved 94% accuracy while maintaining sub-second inference times on mobile devices.
An AI-Powered Multi-Step Meme Generation System
Intelligent meme generator that transforms ideas into culturally relevant memes using a sophisticated multi-step LLM pipeline. Expands user queries into multiple creative variations, integrates real-time knowledge retrieval via Tavily/Perplexity for topical accuracy, and employs in-context learning with few-shot examples to generate contextually perfect meme text.
Key Insight: Implementing a three-stage LLM workflow (query expansion → knowledge retrieval → contextual generation) with few-shot prompting increased meme relevance by 85% and reduced off-topic generations by 70%, creating culturally accurate content at scale.
True expertise is not just claimed; it is validated by the market. I believe in sharing knowledge to elevate the ecosystem, and my technical writing serves as a public record of my thought process and a testament to my ability to communicate complex ideas with clarity and impact.
A Portfolio of Technical Writing & Community Impact
This portfolio of over 65 articles serves as a living document of my exploration and mastery of AI engineering. The core metrics below are key performance indicators that validate the market value of my technical knowledge and communication skills.
A look at the roles, teams, and outcomes where I've scaled AI initiatives from research to production impact.
Architecting and shipping production-grade AI systems while cultivating a global audience through technical thought leadership. Operating at the intersection of LLM engineering, mobile development, and content creation to build products that solve real problems at scale.
Spearheaded strategic pricing intelligence and demand forecasting operations for Fortune 500 pharmaceutical and retail giants, architecting predictive systems that drove millions in revenue optimization. Transformed complex market dynamics into actionable insights, positioning clients ahead of competitive curves.
Pioneered computer vision infrastructure and intelligent automation systems that transformed manual workflows into scalable, API-driven services. Delivered production-ready solutions that exceeded accuracy benchmarks while dramatically accelerating business development operations.
Architected mission-critical data infrastructure and predictive analytics systems for Asia's fastest-growing furniture rental platform, empowering C-suite executives with real-time intelligence and automated KYC processes that scaled customer onboarding operations.
Drove digital transformation initiatives for one of the world's leading engineering software companies, modernizing enterprise workflows through mobile-first integrations and geospatial analytics. Contributed to sustainability research through large-scale environmental data engineering projects.
A strategic blend of research depth and shipping velocity spanning the full AI product stack.
My work is driven by a relentless curiosity to build things that matter. The past 1.5 years as an indie builder has been an intense, end-to-end masterclass in turning complex AI concepts into shipped, revenue-generating products. Now, I am ready to bring that same energy, ownership, and product-focused mindset to a collaborative team. I am actively seeking a role where I can contribute to a larger vision, tackle challenges at a greater scale, and grow alongside the best in the field.