Research papers, benchmark code, presentations, and planning resources from the Anote AI Research Fellowship.
The fellowship goal is to produce publishable research across NLP, RAG, agentic AI, and annotation efficiency, with each paper producing a reusable open-source artifact.
| Area | What is here |
|---|---|
| Active research | 14 current paper projects from the revised 2026 research plan, with owners, fellows, repos, and target venues |
| Paper repos | Standalone anote-ai repositories for each live paper, benchmark, or applied research thread |
| Benchmark code | Experiments for RAG, text classification, question answering, and object detection |
| Research assets | Prior papers, presentations, and video talks for onboarding and background reading |
| Program tracking | Spreadsheets with paper matching, deadlines, venues, owners, and publishing status |
| If you are... | Start with... |
|---|---|
| A fellowship intern | Find your paper in Current Papers, then complete your release packet in researchops/papers/ |
| A researcher reviewing scope | Skim the current paper table to understand the active research portfolio |
| Looking for reusable code | Browse researchcode/ by benchmark area |
| Looking for paper drafts | Open the relevant folder README under researchpapers/ |
| Looking for background material | Review researchpresentations/ and the Video Talks |
See anote_fellowship_tracker.xlsx and the matching spreadsheet for the latest deadlines, venues, owners, and submission progress.
This table reflects the current paper-level organization from the fellowship matching spreadsheet. Several papers now split out of the original 7 track-level starters into standalone repos and venue-specific submissions.
| # | Paper | Repo | Fellow(s) | Track | Owner | Target Venue(s) |
|---|---|---|---|---|---|---|
| 1 | EnterpriseBench: Do Syntactic Tool-Calling Benchmarks Predict Deployment Trustworthiness? | research-enterprisebench | Aye Oyemami | T1a AgenticEval | Spurthi | DAI 2026 Industry Track + AAAI 2027 |
| 2 | Intent Specification as a First-Class Evaluation Object for Tool-Calling Agents | research-intentspecification | Susana Haing | T1a AgenticEval | Spurthi | EACL 2027 Industry Track + AAAI 2027 (reach) |
| 3 | EnterpriseSynth: Agentic SFT + Eval Data from API Schemas Without Live Execution | research-enterprisesynth, Research-Enterprise-Synth-API | Rashmi Thimmaraju | T1b EnterpriseSynth | Spurthi | MLinPL 2026 + AAAI 2027 |
| 4 | RetrievalBench: Cross-Domain Ablation of RAG Techniques over Structured Documents | research-retrievalbench | Hannah Liang, Zirui Han | T5 RetrievalBench | Spurthi | EMNLP AKBC + AAAI 2027 |
| 5 | Financial Document Retrieval: Controlled Ablation of Chunking, Reranking & Metadata | research-financialdocumentretrieval | Elaine Hong, Ebuka Chidubem Uzoama | T5 RetrievalBench | Spurthi | EMNLP FinNLP |
| 6 | Which RAG Techniques Generalize? Legal Contract Retrieval over CUAD | research-raggeneralization | Sindhu Shetty | T5 RetrievalBench | Spurthi | EMNLP NLLP |
| 7 | Does Document Structure Predict Metadata-Annotation Value in RAG? | research-metadataannotation | Victoria Piroian | T5 RetrievalBench | Spurthi | ENCODE 2026 |
| 8 | Failure Propagation in Agentic RAG Pipelines: A Diagnostic Benchmark | research-agenticrag | Lauren Pothuru | T4 RAG Failure | Spurthi | EMNLP ORACLE |
| 9 | AnnotateBench: How Much Labeled Data Do Annotation Strategies Need Across NLP Tasks? | research-annotatebench | Aitong Zhang | T2a AnnotateBench | Spurthi | JDSE 2026 |
| 10 | AnoteCodeBench: Benchmarking Enterprise Code-Generation Agents vs. Claude Code & Codex | research-codebench | Sharon Zheng, Bryant Jiang | T6 CodeEval | Natan | DAI 2026 Industry Track + AAAI 2027 |
| 11 | Orchestrate-Bench: Learning to Choose Tasks, Tools & Code Execution in Multi-Agent Systems | research-orchestratebench | Yidian Chen, Yingzi Gu | T7 Orchestration | Natan | DAI 2026 Industry Track / EMNLP ORACLE + AAAI 2027 |
| 12 | Adversarial Course-of-Action Generation: Game-Theoretic Multi-Agent Algorithms for MEF & GBC | research-coageneration | Arun Kanhai, Alina Kapanova | T8 GameTheory | Natan | DAI 2026 Industry Track + AAAI 2027 |
| 14 | Posture & Sustainment Optimization: Where to Place and How to Sustain Assets (DASH-5) | research-postureandsustainmentoptimization | Amelie Norris, Alyssa Lee | T10 OptDecision | Natan | SAM.gov RFP (DASH-5) |
| 15 | Recursive Self Improvement | TBD | Natan Vidra | TBD | Natan | TBD |
Paper coordination folders under researchpapers/ now use paper-title slugs instead of track codes like T1a or T5. Each folder has a README with the paper topic, GitHub link, YouTube presentation placeholder, paper link, and release-packet link where applicable.
researchpapers/EnterpriseBench/researchpapers/IntentSpecification/researchpapers/EnterpriseSynth/researchpapers/RetrievalBench/researchpapers/FinancialDocumentRetrieval/researchpapers/RAGGeneralization/researchpapers/MetadataAnnotation/researchpapers/AgenticRAG/researchpapers/AnnotateBench/researchpapers/CodeBench/researchpapers/OrchestrateBench/researchpapers/COAGeneration/researchpapers/PostureAndSustainmentOptimization/researchpapers/RecursiveSelfImprovement/
Historical starter drafts that are not in the current paper table are also kept under title slugs:
Research/
├── anote_fellowship_tracker.xlsx # Fellowship tracker and legacy planning sheet
├── main.tex # Reference LaTeX template (RAG paper)
├── researchpapers/
│ ├── EnterpriseBench/ # Paper-title coordination folder
│ ├── IntentSpecification/
│ ├── EnterpriseSynth/
│ ├── RetrievalBench/
│ ├── FinancialDocumentRetrieval/
│ ├── RAGGeneralization/
│ ├── MetadataAnnotation/
│ ├── AgenticRAG/
│ ├── AnnotateBench/
│ ├── CodeBench/
│ ├── OrchestrateBench/
│ ├── COAGeneration/
│ ├── PostureAndSustainmentOptimization/
│ ├── RecursiveSelfImprovement/
│ ├── AgenticEval/ # Historical starter topic
│ ├── AnnotateROI/ # Historical starter topic
│ ├── HumanAITeaming/ # Historical starter topic
│ ├── classification.pdf
│ ├── questionanswering.pdf
│ └── retrieval.pdf # arXiv:2404.07221
├── researchcode/
│ ├── Benchmarking-RAG/
│ ├── Benchmarking-Text-Classification/
│ ├── Benchmarking-Question-Answering/
│ └── Benchmarking-ObjectDetection/
├── researchpresentations/
│ ├── RAG.pdf
│ ├── TextClassification.pdf
│ ├── AI_Talk.pdf
│ └── HumanCenteredAI.pdf
└── researchops/
├── papers/ # Spreadsheet-aligned paper release packets
├── templates/
└── conference-submission-checklist.md
- Read your paper entry in Current Papers and fill in the matching release packet in
researchops/papers/ - Design — write a Research Design Doc and link it in the tracker spreadsheet
- Code — build experiments and link the relevant paper repo, code repo, or benchmark repo
- Results — run experiments, produce tables/figures
- Paper — update the draft in
researchpapers/and/or the linked standalone paper repo - Repo — keep the central repo and the paper-specific repo in sync
- Update — update the tracker spreadsheet, release packet, and publishing assets with all URLs and status
Use researchops/ as the central operating guide for where each research asset should go.
- Papers and code should be centralized in this repo first.
- Paper landing content should be prepared for
anote.ai/research. - Research blog content should be prepared for
anote.ai/blogand Medium. - Research presentations should be prepared for
anote.ai/researchTalks. - Social launch copy should be tracked in the external social spreadsheet.
Start with:
researchops/README.mdresearchops/conference-submission-checklist.mdresearchops/templates/paper-release-packet-template.mdresearchops/templates/intern-outreach-template.mdresearchops/papers/
| Topic | Video |
|---|---|
| Fine Tuning LLMs | YouTube |
| Benchmarking Text Classification | YouTube |
| Benchmarking Q&A Models | YouTube |
| Human Centered AI | YouTube |
| Improving Retrieval for Q&A | YouTube |
| Few Shot Learning Ted Talk | YouTube |