AI / ML Engineer • Builder • Researcher
I work across LLM applications, agentic workflows, NLP, computer vision, and AI systems, while staying grounded in the harder questions around reliability, interpretability, and model behavior.
Pipeline
Signals
About
I come from a computer engineering background, where I started by building software systems and learning how interfaces, APIs, and data flows work in practice. That led naturally into machine learning and AI, especially the parts that sit between model capability and real-world behavior.
My work has ranged from LLM applications and RAG pipelines to agentic workflows, multilingual assistants, and computer vision experiments. I value building systems that are useful in practice, but I am equally interested in understanding why they work, where they break down, and how they can become more reliable and interpretable.
I am still early in this trajectory, but I am increasingly drawn to the harder questions around model behavior, intelligent systems, and the gap between observed performance and actual understanding.
Selected Work
A personalized news companion that combined retrieval, summarization, sentiment analysis, named entity extraction, recommendation, and voice interaction into one research-driven product experience.
Production-oriented AI workflows built around retrieval, context assembly, tool execution, and grounded response generation for real user-facing interactions.
Gesture and vision-driven interfaces built with OpenCV and MediaPipe to explore more natural, real-time ways of interacting with software.
Research / Exploration
Question
How can a personalized AI news system combine retrieval, language understanding, and user-specific context without becoming noisy or unreliable?
Approach
The system retrieved relevant news, extracted entities and sentiment, summarized content for clarity, and designed a user-facing experience around personalization and interaction.
Technologies
NLP, news APIs, summarization, sentiment analysis, named entity recognition, recommendation strategies, and voice-enabled interaction.
What it taught me
A system can be functionally rich and still fail in ways that matter to users: poor retrieval quality, weak personalization signals, and fragile interactions all change how trustworthy the experience feels.
Questions I'm Exploring
How can we understand what representations neural networks actually use?
When should we trust a model's explanation of its own behavior?
How can we distinguish behavioral alignment from underlying mechanisms?
How robust are learned behaviors under distribution shift?
How can AI systems detect and correct their own failures?
How should agentic systems behave when objectives are ambiguous?
Experience
AI / Full-Stack Engineering
I have worked on AI systems that need to function under real product constraints: retrieval quality, response reliability, latency, multilingual behavior, and user trust.
This included building conversational AI experiences with RAG, LangChain, LangGraph, and grounded workflows for website/document interaction, along with multilingual voice and chat interfaces.
Built around
Designed and iterated on RAG pipelines and agentic workflows for conversational AI experiences.
Built multilingual, grounded chatbot systems with retrieval, tool usage, and streaming interactions.
Worked with voice interfaces, including Whisper-based speech-to-text, TTS pipelines, and latency-sensitive conversational flows.
Integrated LangChain and LangGraph-based orchestration workflows with Pinecone, Supabase, Redis, and REST API-backed services.
Worked through real deployment constraints around grounding, hallucination control, latency, and reliability across web and voice interfaces.
Technical Interests
AI / ML
AI Systems
Research
Engineering
Current Questions