We help companies build, fine-tune, evaluate, and deploy AI products.
Fine-tuning LLMs and multimodal models. Custom architectures and training pipelines. From LoRA adapters to full training runs.
Video generation, vision-language systems, and end-to-end inference pipelines orchestrating multiple models in production. VRAM optimization, batch processing, GPU automation.
Retrieval pipeline design, evaluation frameworks, and optimization for production workloads. Evaluation tools and dataset strategies for large-scale corpora.
Autonomous agents that research, reason, and take action. From internal workflow automation to customer-facing agent products with real users.
Literature reviews, feasibility studies, and technical roadmaps grounded in current research. Bridging the gap between papers and production.
Ongoing senior AI guidance without a full-time hire. Roadmaps, hiring plans, architecture decisions, and a trusted partner for board conversations and strategic trade-offs.
Built Friday-VLM by combining Phi-4 with a FastViT-HD vision encoder, plus a full multimodal training pipeline using DeepSpeed and LoRA adapters. Released open-source model artifacts and tooling for reproducible VLM fine-tuning.
Co-authored and implemented behavior-cloning workflows for FLAM, with baseline and training pipeline work across the factored_genie codebase and the LAPO baseline repository.
Read the paper.
Trained a MobileNetv3 CNN on 20K images across 31 food classes, achieving >95% accuracy for real-time inference on a resource-constrained embedded device. Exported to PyTorch Mobile and integrated with the oven's embedded Android OS. End-to-end CV work from data preprocessing and augmentation to model training and optimization.
Developed a multi-agent AI system for conversational interactions with clinical trial datasets (SDTM and ADaM). Integrated GPT-4-32K to dynamically interpret natural language queries into executable code and database queries, enabling real-time data visualization and analysis.
Built and launched an AI agent product that autonomously researches topics daily and delivers personalized briefings with source citations. End-to-end product from agent architecture through subscription billing.
We're Conifer Labs, founded by Kevin Rohling. We love challenging engineering problems and we're happy working across the AI stack: fine-tuning, custom architectures, evaluation frameworks, and inference pipelines. Before founding Conifer Labs, Kevin led AI and engineering teams at startups and has gone from slide deck to product countless times. Kevin holds an MS in Artificial Intelligence from UT Austin (graduated December 2025).
"Exceptionally skilled, committed, and thoughtful. Kevin has the rare ability to deliver complex technology solutions while mentoring team members."Jason Monberg, CEO, Presence Product Group
We publish models, datasets, and research tools on HuggingFace and GitHub. Select projects:
We'd love to hear about your project. Send us a message below or book a call directly.
