AI research + engineering · end-to-end

Bring us your hard problem.
We will deliver the working AI system.

You get serious AI research, practical engineering, and one team from start to handoff.
We work fast, test the whole system, and make sure it works.

Discuss your problem See our work

Solutions

The problem comes first.

The quote lists the test, price, and due dates. You see a working build at each milestone. The handoff includes the code, tests, deployment setup, and notes.

Tell us what is stuck →

YOUR PROBLEM TRAIN / FINE-TUNE ADAPT OPEN SOURCE MODEL APIs COMBINE + BUILD THE FULL SYSTEM METHOD · CHOSEN FOR THE PROBLEM COMPLETE WORKING SOLUTION TESTED TO AGREED ACCEPTANCE CRITERIA
01 · TEST

Evaluate

Set the tests, acceptance criteria, and failure checks before we build.

02 · MODEL

Build or adapt

Fine-tune, adapt open source, use APIs, or combine them to fit the job.

03 · SYSTEM

Integrate

Connect models, data, tools, APIs, and business rules into one usable system.

04 · HANDOFF

Ship and hand off

Receive working software, tests, deployment, documentation, and the repository.

Work

UAV3D photorealistic aerial scene with five drones over a simulated city

Benchmark · NeurIPS 2024

UAV3D, a public large-scale benchmark for 3D perception from drone swarms

Detection, tracking, and collaborative perception from UAV platforms, in the nuScenes format.

  • 1,000 scenes · 500K RGB images · 3.3M annotated 3D boxes
  • Public code, dataset, and leaderboard
H. Ye · first author · NeurIPS 2024 Datasets and Benchmarks Project site →   ▶ Demo video
NovelHive: 200,000 words, 26 voices, 7 minutes
Prior team product work

NovelHive, a consumer AI fiction platform on web and iOS

A production reading platform with web and iOS clients, payments, and multilingual audio.

Team-member prior work · live since 2025 novelhive.ai →
Boston Dynamics Spot running a navigation policy from Frank Liu’s graduate research
Robotics

Sim-to-real navigation on a Boston Dynamics Spot

Frank’s graduate research trained a navigation policy in simulation and demonstrated it on physical hardware.

F. Liu · graduate research, Georgia State · 2024 ▶ Watch the robot →

Research & code

More work you can inspect.

Open the paper, code, or public project behind each card.

ACD-DETR detection results on UAV imagery
Peer-reviewed

ACD-DETR: small-object detection in drone imagery

H. Ye · co-author · Sensors 2025 Read the paper →
MatchXML architecture diagram
Paper + code

MatchXML: classification over millions of labels

H. Ye · first author · IEEE TKDE 2024 Open the code →
DeepPhospho workflow figure from Nature Communications
Paper + code

DeepPhospho: deep learning for phosphoproteomics

F. Liu · co-first author · Nature Communications 2021 Open the code →
WeakNucleiSeg two-branch architecture with nuclei imagery
Paper + code

Nuclei segmentation from point annotations only

F. Liu · first author · ISBI 2022 oral Open the code →
Text-to-image synthesis comparisons on the CUB dataset
Paper + code

Contrastive learning for text-to-image synthesis

H. Ye · first author · BMVC 2021 Open the code →
Taxonomy diagram of LLM attacks and defenses
Survey · preprint

Vulnerabilities and protections in large language models

F. Liu · first author · arXiv 2024 Read the survey →
APLC-XLNet model architecture
Paper + code

Adaptive label clusters for extreme classification

H. Ye · first author · ICML 2020 Read the paper →
GMMC t-SNE feature clusters
Peer-reviewed

Generative Max-Mahalanobis classifiers

H. Ye · co-author · ECML PKDD 2021 Read the paper →
Adversarial reinforcement learning feature selection framework
Peer-reviewed

Adversarial reinforcement learning for domain adaptation

H. Ye · co-author · WACV 2021 Read the paper →
Multi-view 3D detection views on the UAV3D benchmark
Manuscript

PropBEV: temporal query propagation for multi-view 3D detection

H. Ye · manuscript under review · 2026 Manuscript under review

Engagement

01

20-minute call

Start with the problem and the result you need. We will tell you if the job is outside our wheelhouse.

02

Fixed-price SOW

Otherwise, we write a short SOW with the deliverable, fixed price, dates, and acceptance test. SBIR and STTR scopes stay inside the approved plan and award terms.

03

Milestone delivery

At a milestone, you inspect the working code before the invoice goes out. Terms are Net 30. We check data access, citizenship, export controls, and required flow-downs before the project starts.

Team

AI/ML research and production engineering.

Chief Executive Officer

Maya Christine Liu

Maya owns Novela AI outright and runs it. She decides what work we take on and what we turn down, sets the price, and signs the contract. She's the client's point of contact through delivery. Engineering reports to her.

Technical Lead · AI/ML Engineering

Weizhen (Frank) Liu

Frank has spent eight years moving between AI research and production software. DeepPhospho took him into phosphoproteomics; WeakNucleiSeg into medical images; his graduate work put a navigation policy on a Spot robot; NovelHive became a web and iOS product. He holds two master’s degrees in computer science.

  • LLMs, NLP, agents, evaluation, and model adaptation
  • Computer vision, medical imaging, and bioinformatics
  • Robotics, reinforcement learning, and production systems

8 years in AI/ML · Co-first author, Nature Communications & ISBI

AI/ML Research & Engineering

Hui Ye, Ph.D.

Hui worked in software and systems engineering for almost six years before earning his computer science Ph.D. His papers range from extreme text classification to image generation, object detection, and multi-drone 3D perception.

  • Computer vision, detection, tracking, and 3D perception
  • NLP, extreme classification, and generative learning
  • Benchmark, dataset, and production systems engineering

Nearly 6 years in software and systems engineering · NeurIPS, ICML & IEEE TKDE

Tell us what's short-handed.

One paragraph from you. We reply with scope, price, and a delivery date.

We use these details only to respond. Do not submit classified, CUI, export-controlled, confidential, or regulated personal data. See our Privacy Notice.