Shane X. — Senior LLM Engineer

01 — About

Engineer, not evangelist.

I'm an AI/LLM engineer who ships production systems — not demos, not wrappers. My career has been spent at the intersection of NLP, data science, and enterprise software, building tools that real users depend on daily.

At Aunalytics, I led the modernization of legacy deep learning systems into LLM-based architectures for banking clients — building NL2SQL engines, intelligent agent routing, RAG pipelines, and compliance-first safeguards.

Now I focus on LLM fine-tuning (LoRA, DPO, RLHF), agentic workflows (LangGraph, Google ADK), and evaluation frameworks. I publish hands-on walkthroughs on YouTube to share what I learn building.

Years NLP experience

35+

Tutorials published

2K+

YouTube subscribers

95%

Agent eval accuracy

02 — Experience

Where I've built.

Jul 2024 – Present

Independent AI Researcher & Educator

Remote · YouTube @LLMImplementation

Implemented DPO safety alignment on Llama 3.2, achieving 100% jailbreak refusal at ~72% cost reduction vs RLHF/PPO ($28 vs $100+).
Engineered $0.62 prompt-distillation pipeline — 120B teacher → Qwen 30B student via LoRA (15K+ views).
Built fully local RAG agent (LangGraph + Ollama) with ~60% latency reduction versus GPT-4o baseline.
Designed production Career Agent — LangGraph state machine with multi-layer filtering, self-reflection, and 4-layer evaluation framework (95% intent accuracy, 100% SQL correctness).
Delivered paid industry collaboration building end-to-end fine-tuning demos (SFT + RL) using partner SDK.

Jan 2022 – Jul 2024

Data Scientist — NLP / Product Tech Lead

Aunalytics · South Bend, Indiana

NLP Tech Lead for Generative AI — re-architected legacy DL systems into LLM architectures (GPT-4) for enterprise banking clients.
Built NL2SQL engine with vector-based schema pruning (60% token cost reduction) and compliance-first SQL validation.
Engineered Financial Agent with context-aware routing — >95% Precision/Recall, with validation & retry for JSON payloads.
Deployed Support RAG system indexing Jira tickets and logs, reducing triage time by 90% via LLM summarization.
Owned release lifecycle — FastAPI, Docker, Elasticsearch consolidation.

Feb 2020 – Dec 2021

Data Scientist — NLP / NL2SQL

Aunalytics · South Bend, Indiana

Co-authored "An Optimized NL2SQL System for Enterprise Data Mart" for banking data marts.
Built synthetic data generation engine for finance-domain NL–SQL pairs at scale.
Designed semantic value-matching (Elasticsearch) improving exact-match accuracy from ~1% to ~45%.
Re-architected NL autocomplete to low-latency Elasticsearch service (seconds → milliseconds).

Jul – Dec 2019

Data Research & Analysis Intern

Impending Bloom · New York

Built ML classification and entity-matching pipelines (BigQuery, PostgreSQL, Elasticsearch, MongoDB) achieving ~90% precision.

03 — Selected Projects

What I've shipped.

Production · Banking

Enterprise NL2SQL Engine

Context-aware LLM-based natural language to SQL engine for community bank analysts. Vector-based schema pruning (60% token savings), business term disambiguation, compliance-first SQL validation with whitelisting, and a self-improving few-shot example store.

GPT-4PythonElasticsearchFastAPIDockerSentence-BERT

Production · Banking

Financial Agent Engine

Multi-tool analytics agent with intelligent routing for banking workflows. Context-aware tool selection (>95% Precision/Recall), JSON payload validation with retry, and dynamic context injection. Designed for auto-approval and human escalation flows.

LLM AgentsTool RoutingPythonJSON ValidationFastAPI

Automation

Career Agent — Daily Pipeline

AI job matching system in daily production. LangGraph state machine with ReAct self-reflection, multi-layer filtering (heuristic + LLM), artifact generation, and 4-layer evaluation framework (95% intent accuracy).

LangGraphGeminiSQLiteMilvusFastAPI

Alignment

DPO Safety — Llama 3.2

Compared PPO-RLHF vs DPO for safety alignment. DPO achieved 100% jailbreak refusal at ~72% lower cost. Custom eval harness with LLM-as-a-judge rubric and adversarial testing.

DPORLHFLoRAUnslothEvaluation

RAG

Local RAG Agent — Zero Cost

Fully local RAG with LangGraph + Ollama and self-correcting retrieval. EmbeddingGemma with dimension truncation (768→256). 1.3s latency — 60% faster than GPT-4o.

LangGraphOllamasqlite-vecEmbeddingGemma

Fine-tuning

$0.62 Prompt Distillation

Knowledge distillation from 120B teacher to Qwen 30B MoE student via LoRA on Tinker. Synthetic data generation + distributed training for under $1. Tutorial reached 15K+ views.

TinkerLoRAQwen 30B MoEDistillation

Evaluation

LLM Evaluation Pipeline

Perplexity diagnostics, Functional Correctness (pass@k), Semantic Similarity (embeddings), and LLM-as-a-Judge with position bias testing. Three-pillar prompt design for robust scoring.

Pythonsentence-transformersHugging FaceOpenRouter

05 — Technical Stack

What I work with.

AI & Agents

LangGraph / LangChain
Google ADK / MCP
RAG Pipelines
Tool Routing & Orchestration
LLM Evaluation (AI-as-Judge)
Ollama / LiteLLM

Fine-tuning & Training

PyTorch / Hugging Face
Unsloth / QLoRA / LoRA
DPO / PPO / RLHF
Prompt Distillation (Tinker)
4-bit Quantization
Weights & Biases

MLOps & Infra

Python / FastAPI
Docker / CI/CD
GCP (Vertex AI, Cloud Run)
LangSmith / Langfuse
SQL / Elasticsearch
Vector DBs (Milvus, sqlite-vec)

Models

GPT-4 / GPT-OSS
Gemini 2.5 / 3
Claude
Llama 3 / Qwen / DeepSeek
IBM Granite
Sentence-BERT / EmbeddingGemma

Data Science

NL2SQL / Semantic Parsing
Statistical Modeling
scikit-learn / XGBoost
Pandas / NumPy
Experiment Design
Semantic Search

Workflow

Git / GitHub
uv / pip
Google Colab
VS Code / Cline
LangGraph Studio
macOS / Linux

I build production AI
for enterprise.

LLM Implementation

I build production AIfor enterprise.

LLM Implementation

I build production AI
for enterprise.