.webp)
Resumes
August 13, 2026
Resume Example
Real-world LLM engineer resume examples across LLM engineering, LLM specialist, and LLM pipeline developer roles, with the accuracy, cost, reliability, and evaluation metrics that applied AI teams look for.
Ishaan Mehta
(415) 555-0149 ◇ San Francisco, CA
ishaan.mehta@example.com ◇ linkedin.com/in/ishaan-mehta ◇ ishaanmehta.com
Objective
LLM engineer with 5 years building production systems on large language models, currently the lead LLM engineer for a document intelligence product processing 40 million pages a month for 600 enterprise customers. Built the retrieval-augmented generation platform that raised answer accuracy from 71% to 91% on a 5,000-question benchmark, cut inference cost per query 64% through model routing, caching, and fine-tuning, and holds a 99.95% availability record on the inference service over 2 years. Fine-tuned 6 models that beat the general-purpose baseline by an average of 14 points on domain tasks. Builds language model systems that are measured, monitored, and cheap enough to run at scale.
Education
M.S. in Computer Science, Machine Learning, Golden Gate Crown University 2018 – 2020
Thesis on efficient fine-tuning of transformer models. Coursework in Deep Learning, Natural Language Processing, Distributed Systems, and Information Retrieval.
B.Tech in Computer Science and Engineering, Bay Line Institute of Technology 2014 – 2018
Graduated with distinction. Coursework in Algorithms, Machine Learning, Databases, and Operating Systems. Also holds an AWS Certified Machine Learning Specialty certification earned in 2022.
Skills
LLM Engineering
Retrieval-augmented generation architecture, embedding and vector search design, chunking and indexing strategies, prompt and context management, tool calling and agent orchestration, fine-tuning with parameter-efficient methods, model evaluation and selection
Inference and Infrastructure
Inference serving and autoscaling, model routing and caching, latency and cost optimisation, quantisation and batching, observability for language model systems, guardrails and safety filtering, GPU infrastructure
Engineering Practice
Python and Go, distributed systems design, evaluation harnesses and offline benchmarks, A/B experimentation, data pipelines for training and evaluation, technical leadership and mentoring, design documentation
Tools & Platforms
PyTorch, Hugging Face Transformers and PEFT, vLLM, LangGraph, Claude and open-weight model APIs, Pinecone and pgvector, Kubernetes, AWS and Ray, Weights and Biases, LangSmith, Terraform
Experience
Lead LLM Engineer 01/2023 – Present
Golden Gate Crown AI San Francisco, CA
- Lead LLM engineering for a document intelligence product processing 40 million pages a month for 600 enterprise customers, leading a team of 7 engineers.
- Built the retrieval-augmented generation platform with hybrid search, a reranker, and structured context assembly that raised answer accuracy from 71% to 91% on a 5,000-question benchmark and cut unsupported answers 80%.
- Cut inference cost per query 64% through model routing across 3 model tiers, semantic caching with a 38% hit rate, and 6 fine-tuned models, and hold a 99.95% availability record on the inference service over 2 years at 20 million queries a month.
Machine Learning Engineer, NLP 07/2020 – 12/2022
Bay Line Search Technologies Palo Alto, CA
- Built semantic search and question answering systems for an enterprise search product used by 200 customers over 2.5 years.
- Shipped the first language model based answer feature in 2022, reaching 78% accuracy on a 2,000-question benchmark and a 30% click-through lift.
- Built a training pipeline that fine-tuned embedding models on 4 million query pairs, raising retrieval recall at 10 from 62% to 84%.
Machine Learning Research Intern 06/2019 – 09/2019
Bay Line Search Technologies Palo Alto, CA
- Built a passage reranking model during a 14-week internship that raised top-1 accuracy 18% on an internal benchmark.
- Wrote an evaluation harness for 3 retrieval models used by the team for 2 years.
- Received a full-time offer at the end of the 14-week internship based on the reranking model.
Projects
Retrieval-Augmented Generation Platform. Designed a platform with layout-aware chunking for 40 million pages a month, hybrid dense and keyword retrieval over 2 billion chunks, a fine-tuned cross-encoder reranker, structured context assembly with citation tracking, and an evaluation harness of 5,000 questions across 12 document types, which raised answer accuracy from 71% to 91%, cut unsupported answers 80%, and serves 20 million queries a month.
Inference Cost Programme. Built a model routing layer classifying queries into 3 complexity tiers with a fine-tuned 7-billion parameter model handling 55% of traffic, semantic caching with a 38% hit rate, continuous batching on vLLM, and quantisation for the smaller tiers, validated against the full benchmark, which cut cost per query 64%, saving about $4.8M a year, with a 1-point accuracy change.
Domain Fine-Tuning Programme. Fine-tuned 6 open-weight models with parameter-efficient methods on 400,000 curated examples across extraction, classification, and summarisation tasks, with a data pipeline for labelling, deduplication, and quality filtering, which beat the general-purpose baseline by an average of 14 points on domain benchmarks and cut latency 60% on those tasks.
Extra-Curricular Activities
- Speaker at 4 machine learning conferences on retrieval-augmented generation and inference cost.
- Maintain an open-source evaluation library for retrieval systems with about 2,500 stars on GitHub.
- Volunteer mentor for a machine learning fellowship, coaching about 4 fellows a year.
Leadership
- Lead an LLM engineering team of 7 with design reviews, roadmap ownership, and hiring, with 2 promotions in 2 years.
- Own the inference platform budget of $7M a year and present cost, accuracy, and reliability metrics to the chief technology officer monthly.
- Wrote the company standard for language model evaluation and release adopted by 4 product teams.
.webp)
%20Which%20Should%20You%20Use.webp)
