NEARSHORE LLM ENGINEERS
Model-layer specialists from Costa Rica and Colombia who fine-tune, quantize, serve, and evaluate large language models in production — not just call an API. Full US timezone overlap, LoRA/QLoRA and vLLM depth, at 40-60% below domestic rates with a 90-day replacement guarantee.
Model-Layer Specialists.
Not Just API Callers.
Costa Rica and Colombia produce LLM engineers with 4-8 years of experience fine-tuning open-weight models, optimizing inference costs, and running production evaluation harnesses. They operate within 0-2 hours of US time zones and cost 40-60% less than a domestic equivalent.
A nearshore LLM engineer is a specialist based in a geographically close country (Costa Rica or Colombia for US companies) who works at the model layer itself — fine-tuning foundation models with LoRA and QLoRA, quantizing models for production serving, optimizing inference latency and token cost, and building systematic evaluation harnesses to catch regressions before they ship. This is distinct from an AI engineer, who typically builds the application layer on top of a model (RAG pipelines, agentic workflows, product integrations). Kore BPO sources, vets, and places LLM engineers so you skip months of domestic recruiting for a genuinely scarce specialization.
We Screen for the Model Layer, Not the Prompt Layer.
Most “AI engineer” candidates can wire an LLM API into a product. Far fewer can explain why a quantized model’s output degraded, tune a LoRA adapter without catastrophic forgetting, or diagnose why inference latency spiked after a batching change. Our technical screen filters specifically for that depth.
- ✓ Technical screen: LoRA/QLoRA fine-tuning, PEFT, quantization (GGUF, AWQ, GPTQ), and dataset curation depth
- ✓ Inference serving depth: vLLM, TGI, or TensorRT-LLM matched to your latency and throughput targets
- ✓ English communication check: written and spoken fluency at production incident level
- ✓ Evaluation discipline: regression harnesses, offline eval sets, and hallucination/safety scoring, not vibes-based review
What Your LLM Engineer Will Know
Kore BPO screens for model-layer depth across fine-tuning, serving, and evaluation — the specific slice of AI engineering that most generalist candidates cannot cover.
Fine-Tuning & Adaptation
LoRA, QLoRA, and full-parameter fine-tuning with PEFT and Hugging Face Trainer, plus instruction tuning and RLHF/DPO alignment workflows.
Inference & Serving
vLLM, Text Generation Inference (TGI), TensorRT-LLM, and Triton Inference Server for high-throughput, low-latency model serving.
Quantization
GGUF, AWQ, GPTQ, and bitsandbytes for 4-bit and 8-bit quantization, balancing model size, latency, and output quality.
Evaluation & Observability
promptfoo, RAGAS, LangSmith, and Arize Phoenix for regression testing, hallucination scoring, and production LLM monitoring.
Vector & Embedding Infra
Custom embedding model training and evaluation, plus Pinecone, Weaviate, and Qdrant integration at production scale.
Model Providers & Open-Weight
OpenAI, Anthropic, and Gemini API depth, plus open-weight model deployment: Llama, Mistral, and Qwen via Hugging Face.
Cost & Latency Optimization
Token budget management, response caching, dynamic batching, and model routing across providers to hit cost-per-request targets.
Safety & Guardrails
Content filtering, jailbreak and prompt-injection testing, PII redaction, and rate-limiting for production-facing model endpoints.
From Discovery to First Commit in 3 Weeks
Our placement process is built around your model stack, not generic technical benchmarks. You get engineers who are pre-matched to your frameworks and production environment.
Discovery Call
We map your model stack, use cases (fine-tuning, serving, evaluation), and the frameworks and hardware your engineer will work in from day one.
Candidate Matching
Within 72 hours, we present 2-3 pre-vetted profiles. Each has passed our async assessment covering fine-tuning judgment, quantization trade-offs, and inference debugging.
Your Interview
You run a 60-90 minute technical session focused on your real model problem. We provide a structured interview guide tailored to your stack if needed.
Offer & Start
We handle Latin American employment contracts, payroll, benefits, and HR administration. The engineer starts on your agreed date and joins your sprint from day one.
90-Day Guarantee
If the placement does not meet expectations on technical skills or fit within 90 days, we re-run the full search at no additional cost.
Nearshore LLM Engineer Rates vs US Rates
All-in costs include salary, payroll taxes, benefits, and Kore BPO account management. No upfront placement fees.
| Experience Level | US Market Rate | Nearshore via Kore BPO | Annual Savings |
|---|---|---|---|
| Mid-Level (3-5 yrs, fine-tuning + serving) | $130,000 – $160,000 | $65,000 – $85,000 | $65,000 – $75,000 |
| Senior (5-8 yrs, LLMOps + inference optimization) | $160,000 – $210,000 | $85,000 – $115,000 | $75,000 – $95,000 |
| Staff / Principal (8+ yrs, model platform architecture) | $210,000 – $290,000 | $115,000 – $145,000 | $95,000 – $145,000 |
Rates are all-in annual figures through Kore BPO covering Costa Rica and Colombia placements. See full rate breakdown in the Nearshore LLM Engineers Salary Guide.
Nearshore LLM Hiring Works When
✓ Good Fit
- ✓ Your AI product’s bottleneck is model quality, cost, or latency — not the application layer
- ✓ You need to fine-tune or quantize an open-weight model for a specific domain
- ✓ You are self-hosting inference and need to control token cost and latency directly
- ✓ Real-time US timezone collaboration is required
- ✓ Cost savings of 40-60% matter to your unit economics
✕ Less Ideal
- ✕ You only need to call an LLM API from your product — that is an AI engineer hire
- ✕ You need a short-term freelancer for a one-week prompt experiment
- ✕ Your use case requires classified government clearance
- ✕ No internal model infrastructure or evaluation baseline exists yet
Nearshore LLM Engineers: Common Questions
What’s the difference between an LLM engineer and an AI engineer?
An AI engineer typically builds the application layer on top of a model: RAG pipelines, agentic workflows, and product integrations using frameworks like LangChain or LlamaIndex. An LLM engineer works one layer deeper, at the model itself — fine-tuning with LoRA or QLoRA, quantizing for production serving, optimizing inference cost and latency, and building the evaluation harnesses that catch regressions. Many teams need both roles; some AI product bottlenecks can only be fixed by someone who can touch the model directly, which is when an LLM engineer becomes the right hire over an AI engineer. See our Nearshore AI Engineers page if your need is application-layer.
How quickly can you place a nearshore LLM engineer?
With Kore BPO, the typical timeline is 10 to 14 business days from discovery call to first candidate presentation. You receive 2 to 3 fully-vetted profiles with video introductions and async technical assessment results. Your interview and offer process adds 3 to 5 business days in most cases, putting the engineer contributing to your model pipeline within three weeks of starting the search.
What fine-tuning and serving frameworks will the engineer know?
Our LLM engineers come with production depth in LoRA, QLoRA, and PEFT for fine-tuning, plus GGUF, AWQ, and GPTQ for quantization. On the serving side, we screen for vLLM, TGI, and TensorRT-LLM experience. We match specifically to your stack during the discovery call rather than sending generalist profiles. If your environment uses less common tooling, we discuss candidly whether a 2-3 week skills ramp is realistic before starting the search.
Will a nearshore LLM engineer work US business hours?
Yes. Costa Rica (UTC-6) and Colombia (UTC-5) operate within 0 to 2 hours of US Eastern and Central time zones. Your LLM engineer can join sprint planning, model review sessions, production incidents, and evaluation reviews in real time without overnight shifts. This real-time collaboration is a core reason US companies choose Latin America over Southeast Asian alternatives for embedded model-layer roles.
What happens if the LLM engineer placement does not work out?
Kore BPO backs every placement with a 90-day replacement guarantee. If the engineer does not meet your expectations for technical skills or performance within the first 90 days, we re-run the full search and placement at no additional cost. The guarantee covers both technical mismatches and soft-skill or cultural fit issues documented in writing between your team and your Kore BPO account manager.
What US Teams Say About Nearshore LLM Hiring
Our Costa Rica LLM engineer took our fine-tuning pipeline from a one-off notebook experiment to a repeatable LoRA training process with real eval gates. Inference cost per request dropped by a third after the quantization work. Hired through Kore BPO in 11 days.
We needed someone who could actually debug why our self-hosted model’s latency spiked under load, not just tune prompts. The Kore BPO candidate found a batching misconfiguration in our vLLM setup in the first week and had a fix shipped before our next release.
The eval harness our LLM engineer built catches regressions before they reach customers, which we genuinely did not have before. The timezone overlap with Colombia means we review results together the same morning, not the next day.
Also Hiring in This Space?
Learn More About Nearshore LLM Engineering
BUILD YOUR NEARSHORE LLM TEAM
Get pre-screened LLM engineers from Costa Rica and Colombia on your desk within 72 hours.
GET STARTED TODAYNo upfront fees | 90-day replacement guarantee