Services

Fine-tuning and post-training services, end to end.

Six disciplines, one team. Buy them as a complete production engagement, or pick the piece your team is missing.

Supervised fine-tuning

Teach a base model your formats, terminology and procedures from curated demonstrations. Full-parameter, LoRA, QLoRA and continued pre-training.

from $48k
3–6 weeks 02

RLHF, DPO & preference tuning

Align judgment and style with expert preferences, verifiable rewards and reward models. DPO, KTO, ORPO, GRPO and PPO.

from $72k
4–8 weeks 03

Model distillation

Transfer a frontier model’s behavior on your task into a small, cheap, private student model.

from $56k
3–6 weeks 04

LLM evaluation

Custom eval suites, calibrated LLM-as-judge, red-teaming and release gates you can run in CI.

from $24k
2–4 weeks 05

Training data

Data strategy, log mining, expert annotation, preference collection and filtered synthetic data.

scoped
per program 06

Deployment & model ops

Quantize, serve, monitor and retrain. We run it, or we hand your platform team a clean runbook.

from $18k/mo
retainer

Stack

We bring opinions, not lock-in.

We work in your cloud and with your tools. These are the ones we use most often.

Clouds

AWS, Google Cloud, Microsoft Azure, Oracle Cloud, CoreWeave, Lambda, Crusoe, on-premises clusters

Data platforms

Databricks, Snowflake, BigQuery, S3/GCS/ADLS lakehouses, Postgres

Base models

Llama, Qwen, Mistral, Gemma, Phi, DeepSeek, and hosted fine-tuning APIs from the major labs

Training

PyTorch FSDP, DeepSpeed, Megatron, TRL, Axolotl, torchtune, verl, custom RL loops

Serving

vLLM, SGLang, TensorRT-LLM, Triton, llama.cpp for edge, managed endpoints on every major cloud

Evaluation

Our own harness, plus lm-eval-harness, Inspect, promptfoo and your existing observability stack

FAQ

Buying services, not software

Do you sell a platform or software licence?

No. We are a services firm. You get the model, the code and the eval harness; you are free to run them on any infrastructure or platform you already use.

Can you work with our in-house ML team?

Most of our engagements are joint. We embed with your team, pair on the training code and leave them able to run the next retraining cycle without us.

Can we buy evaluation or data services on their own?

Yes. Eval suites and data programs are often the first engagement — they make every later model decision cheaper and faster.

Bring us the task. We'll bring the model.

A 45-minute technical briefing with a senior research engineer. You leave with a feasibility read, a data plan and a fixed quote — whether or not you hire us.

Book a technical briefing →