LLM Evaluation | RAG Evaluation | Voice AI
I work on practical AI systems where model behavior matters: RLHF-style response evaluation, retriever comparison, grounded answers, voice-agent edge cases, and local-first LLM applications.
Profile
My background combines LLM evaluation work, production voice AI integration, and business-facing product experience. I care about whether a model followed the instruction, whether a RAG answer is grounded in evidence, and whether a voice pipeline can handle realistic failures such as silence, unclear transcripts, and interruptions.
I am currently targeting remote or relocation-friendly roles in AI evaluation, RLHF, RAG and GenAI applications, voice AI, conversational AI, and applied AI engineering.
Selected public work
Config-driven workbench for pairwise response scoring, RLHF-style rubrics, JSONL validation, multi-annotator agreement, disagreement cases, failure tags, and Markdown/JSON reports.
Measures retrieval recall, citation coverage, required-fact coverage, grounded answer rate, retriever ranking behavior, and failure tags over a small document corpus.
Local voice-agent framework with STT, assistant, and TTS provider boundaries, streaming-style artifacts, p50/p95 latency traces, batch scenarios, and reports.
Focus areas
Prompt design, side-by-side response review, rubric-based scoring, preference data, coding/data-science evaluation, ground-truth corrections, annotator agreement, and failure-mode analysis.
Document ingestion, chunking, retrieval, retriever comparison, citations, groundedness checks, local LLM apps, LangChain, FAISS, Ollama, Streamlit, and Gradio workflows.
STT/TTS integration, provider boundaries, latency tracing, intent routing, batch scenario testing, interruption handling, and LLM-backed assistant quality review.
Experience
Integrated TTS and STT models into PreCall AI and contributed to Turing-sourced RLHF pipelines for xAI and Meta Llama 4, evaluating model behavior across Python and data-science tasks.
Built customer discovery, communication, and business-context judgment through high-volume consultative education sales.
Resume variants
Available for
Best fit: AI evaluation, RLHF, RAG/GenAI applications, voice AI, conversational AI, and Python-based applied AI systems.