Consulting
I am a Principal AI Engineer who builds LLM systems end to end: retrieval and search, fine-tuning on private data, and the multi-node training and serving infrastructure underneath. I take on a small number of consulting engagements where that experience shortens the path from prototype to something you can rely on.
The write-ups on this site are the portfolio. If the problems in them look like yours, we should talk.
Four shapes of engagement. Each starts with a free call and a written proposal with a fixed price. If your problem does not fit any of these, describe it anyway.
Fixed scope · about two weeks
A written assessment of the system you have or the one you are planning: retrieval design, evaluation, cost and latency, failure modes, and the order to fix them in. You leave with a prioritized plan your team can execute with or without me.
Fit: Best when a prototype works in a demo and nobody is sure it will hold up with real users.
Scoped project · typically four to eight weeks
Hybrid search (BM25 plus dense vectors plus reranking), agentic RAG that routes queries to specialized retrievers, and the evaluation harness that tells you whether a change helped. Built on your stack, handed over with documentation.
Fit: Best when answers are inconsistent, retrieval quality is guesswork, or a specialized domain such as healthcare needs more than vanilla RAG.
Scoped project
Supervised fine-tuning and DPO on open-weight models (Qwen, LLaMA, DeepSeek and others), multi-node training with DeepSpeed, torchrun or Ray, plus the checkpointing, NCCL networking and observability that keep a long run alive.
Fit: Best when prompting has hit its ceiling, inference costs need to come down, or your data cannot leave your infrastructure.
Monthly retainer
A standing weekly working session plus asynchronous review of designs, pull requests and vendor decisions. Senior judgement for a team that is building AI features without a senior AI engineer on staff.
Fit: Best for a small engineering team that needs direction more than headcount.
Each of these links to a technical write-up on this site. The training runs include the parts that went wrong. I would rather you judge the work than a testimonial.
Multi-node BF16 training runs of the 671B-parameter DeepSeek-V3 mixture-of-experts model across three and four A100 nodes: DeepSpeed and Ray Train orchestration, NCCL networking, driver and fabric-manager stability, checkpoint hygiene and Prometheus/Grafana monitoring. Written up as an eight-part series plus a technical overview of the model.
Supervised fine-tuning with ZeRO-3 CPU offload and LoRA, followed by Direct Preference Optimization on preference data, for a content-generation workload.
How I approach query routing to specialized agents over a lexical-plus-semantic retrieval stack with neural reranking and diversity optimization, plus the data pipelines that feed it. The architecture I would start from for a domain-specific search product, written up in full.
Architecture and lessons from building Bayesian MMM as a product rather than a PDF: asynchronous GPU training, budget optimization, and a UI that turns posteriors into recommendations non-technical stakeholders can act on.
Sixteen-GPU fine-tuning with DeepSpeed ZeRO-3 offload and LoRA, with stable loss curves and efficient memory use.
Fine-tuning and serving diffusion models for image generation workloads, including the technical details most write-ups skip.
You describe the problem; I ask the questions that decide whether I can help. No charge, no pitch deck.
Scope, deliverables, timeline and a fixed price, in writing, before any work starts. If a smaller engagement would get you there, I say so.
Working code in your repositories, a weekly demo, and documentation your team can run with after I leave.
I am Thomas Kalnik, a Principal AI Engineer and programming generalist. I have built AI systems in healthcare, digital marketing and the creator economy. Before that I worked in data engineering and data science at Criteo, 1upHealth, Dynamo Software and Wellington Management. That background matters: most production AI problems turn out to be data and infrastructure problems wearing a model costume.
I hold a Master's in Computer Science from the University of Illinois Urbana-Champaign, an MBA and MS in Finance from Northeastern University, and a Bachelor's from SUNY Geneseo. The finance training shows up in how I scope work: I care about the cost of a system and the business outcome it is supposed to move, not just whether the model is clever.
More on the about page, and in the technical posts.
Tell me what you are building and where it is stuck. I reply personally within two business days, usually with a few questions and a suggested next step. Prefer email? kalnik.thomas@gmail.com.