[
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Prof. Mara Ellison — Profile",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "# Prof. Mara Ellison — Profile\n\n> Illustrative sample content for the Lab Assistant RAG template. The lab, people,\n> systems, and metrics below are **fictional** and exist only to demonstrate the app.",
  "chunk_id": 0
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Brief description (at a glance)",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Brief description (at a glance)\nDr. Mara Ellison is an Associate Professor of Computer Science at Northgate University\nwho builds **scalable and trustworthy machine learning for structured and\nknowledge-intensive problems**. Her group works across **graph machine learning**,\n**retrieval-augmented and agentic large language models**, **citation-graph retrieval**,\nand **cost-efficient LLM systems**. Her guiding philosophy: *\"Structure is a signal —\nmost hard problems get easier once you model the relationships, not just the items.\"*",
  "chunk_id": 1
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Who she is",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Who she is\nDr. Mara Ellison directs the **Structured Intelligence Lab** in the **Department of\nComputer Science at Northgate University**. She earned her Ph.D. in Computer Science in\n2016 and has published widely on graph learning and retrieval-augmented generation. Her\nwork is supported by public research grants and industry gifts.\n\nHomepage: https://www.example.edu/~mellison/\nGoogle Scholar: https://scholar.google.com/citations?user=ELExample01&hl=en",
  "chunk_id": 2
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Research mission",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Research mission\nThe lab builds **scalable and trustworthy machine learning for structured domains**. A\nrecurring theme: turn a messy real-world question into a problem over *relationships* —\na graph, a citation network, a retrieval structure — and then make the model reason over\nthat structure faithfully and cheaply.\n\n### Core research areas\n- **Graph neural networks & graph learning** — text-attributed graphs, graph encoders,\n  and faithful GNN explanations.\n- **Retrieval-augmented generation (RAG)** — graph-aware retrieval (GraphWeave),\n  citation-graph retrieval for scholarly QA (CiteTrace), and spatial RAG (GeoReason).\n- **Efficient & cost-aware LLM systems** — routing between cheap and expensive models\n  (ThriftRoute) and online distillation for cheaper serving.\n- **Trustworthy AI** — grounding, uncertainty, and explanation for retrieval and graph\n  models.",
  "chunk_id": 3
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Research impact & reach (illustrative)",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Research impact & reach (illustrative)\n- The lab is best known for **GraphWeave**, which extends RAG from flat text chunks to\n  relevant subgraphs, and **CiteTrace**, which retrieves over citation links for\n  better-grounded literature answers.\n- **ThriftRoute** reduces expensive large-model API calls through selective invocation\n  and online distillation — the basis for the cost-aware routing used inside this very\n  assistant.\n- **LitMap** treats literature discovery as a graph-navigation problem rather than a\n  ranked list.",
  "chunk_id": 4
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Selected recognition (illustrative)",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Selected recognition (illustrative)\n- Best Paper Runner-up at a top information-retrieval venue for **CiteTrace**.\n- Early-career research award for work on graph-aware retrieval.",
  "chunk_id": 5
 },
 {
  "doc_id": "profile",
  "title": "Profile of Mara Ellison — Current Ph.D. students and their topics (fictional)",
  "source": "profile",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Current Ph.D. students and their topics (fictional)\n- **Ana Ferreira** — text-attributed graphs, graph encoders, literature-graph discovery (LitMap).\n- **Kian Roswell** — graph & citation-graph RAG, scholarly question answering (GraphWeave, CiteTrace).\n- **Priya Nandakumar** — spatial reasoning and spatial retrieval (GeoReason).\n- **Tomas Beck** — LLM efficiency, cost-aware routing, distillation (ThriftRoute).\n- **Sofia Marchetti** — GNN explainability and faithful explanations.",
  "chunk_id": 6
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Frequently Asked Questions — Prospective Students & Collaborators",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "# Frequently Asked Questions — Prospective Students & Collaborators\n\n> Illustrative sample content for the Lab Assistant RAG template (fictional lab).",
  "chunk_id": 7
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Is the lab taking new Ph.D. students?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Is the lab taking new Ph.D. students?\nThe Structured Intelligence Lab grows most years and regularly considers strong\napplicants. Admission is through the **Northgate University Computer Science Ph.D.\nprogram** — you apply to the department and indicate interest in the lab's areas.\nApplicants who already understand the lab's research (graph learning, RAG/agentic LLMs,\ncitation-graph retrieval, efficient LLM systems) and can point to relevant projects or\ncoursework tend to stand out.",
  "chunk_id": 8
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — What does Prof. Ellison look for in students?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## What does Prof. Ellison look for in students?\nStrong fundamentals in machine learning and mathematics, the ability to turn a real\nproblem into a structured learning problem, solid programming, and genuine curiosity.\nEvidence of initiative — a paper, an open-source project, a thoughtful re-implementation\nof recent work — speaks louder than a résumé line.",
  "chunk_id": 9
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — How should I reach out?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## How should I reach out?\nEmail is best. A good cold email is short and specific: name the paper or research\ndirection you care about, say in one or two sentences how your background connects, and\nattach a CV. Generic mass emails rarely land. You can also leave your details with this\nassistant (below) and they will be passed along.",
  "chunk_id": 10
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Can the lab supervise undergraduate / master's research or internships?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Can the lab supervise undergraduate / master's research or internships?\nYes — the lab works with motivated students at multiple levels. The same advice applies:\ncome in having read some of the lab's work and with a concrete interest.",
  "chunk_id": 11
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — I'd like to collaborate (academia or industry). Is that possible?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## I'd like to collaborate (academia or industry). Is that possible?\nYes. The lab's work on graph learning, retrieval, and efficient LLM systems applies\nacross many domains. Describe the problem, the data, and what a successful collaboration\nwould look like.",
  "chunk_id": 12
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Where can I learn more about the research?",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Where can I learn more about the research?\n- Homepage: https://www.example.edu/~mellison/\n- Google Scholar: https://scholar.google.com/citations?user=ELExample01&hl=en\nAsk this assistant about a specific topic (e.g. \"GraphWeave\", \"citation-graph RAG\",\n\"GNN explainability\", \"cost-aware routing\") for a cited summary.",
  "chunk_id": 13
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Disclaimer",
  "source": "faq",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## Disclaimer\nThis is an AI assistant prototype that answers questions about the lab's research from\nits knowledge base. Responses are generated automatically and may be incomplete or\nimperfect; they do not constitute official statements from the lab or the university.\nAlways confirm important details (deadlines, admissions, funding) through official\nchannels.",
  "chunk_id": 14
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — Lab Tools & Systems",
  "source": "tools",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "# Lab Tools & Systems\n\n> Illustrative sample content for the Lab Assistant RAG template (fictional systems).",
  "chunk_id": 15
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — LitMap — graph-based literature discovery",
  "source": "tools",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## LitMap — graph-based literature discovery\nA system that helps researchers map a field rather than just rank hits. Instead of\nreturning a ranked list like a standard search engine, LitMap is **coverage-oriented**:\nit builds a graph over the literature so a user can see the structure of a research area,\ndiscover related work, and navigate connections between papers and ideas. It reflects the\nlab's broader thesis that **graph structure is a first-class signal** for understanding\nand retrieval.",
  "chunk_id": 16
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — ThriftRoute — cutting LLM API cost through routing + distillation",
  "source": "tools",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## ThriftRoute — cutting LLM API cost through routing + distillation\nA method/toolkit that reduces the number of expensive large-LLM API calls through\n**selective invocation** (only calling the large model when it is actually needed)\ncombined with **online distillation** (learning from the large model's outputs so a\ncheaper model can handle more cases over time). This is the research basis for the\n**cost-aware model routing** used inside this very assistant: cheap questions are\nanswered by a small/fast model and only substantive research questions escalate to a\nstronger model.",
  "chunk_id": 17
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — How this assistant embodies the lab's research",
  "source": "tools",
  "url": "https://www.example.edu/~mellison/",
  "topics": [],
  "authors": [],
  "text": "## How this assistant embodies the lab's research\nThis lab assistant is itself a small demonstration of several of the lab's lines of work:\n- **Retrieval-augmented generation (RAG)** over the lab's publications — see **GraphWeave**.\n- **Citation-/graph-aware retrieval**: after vector search, results are expanded one hop\n  along a paper graph (shared authors and topics), echoing **GraphWeave** and **CiteTrace**.\n- **Trustworthy generation**: answers are grounded in retrieved sources with inline\n  citations, and the assistant declines to answer when the corpus does not support a\n  claim — reflecting the lab's emphasis on robustness and uncertainty.\n- **Cost-aware routing**: a direct nod to **ThriftRoute**.",
  "chunk_id": 18
 },
 {
  "doc_id": "graphweave",
  "title": "GraphWeave: Graph Retrieval-Augmented Generation over Document Subgraphs",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/graphweave",
  "topics": [
   "rag",
   "graph",
   "llm"
  ],
  "authors": [
   "Kian Roswell",
   "Ana Ferreira",
   "Mara Ellison"
  ],
  "text": "GraphWeave: Graph Retrieval-Augmented Generation over Document Subgraphs (NAACL Findings 2025). Authors: Kian Roswell, Ana Ferreira, Mara Ellison. Extends retrieval-augmented generation from retrieving isolated text chunks to retrieving relevant subgraphs, so the language model can reason over the structure and relationships among retrieved documents instead of a flat list. GraphWeave improves multi-hop answer faithfulness on knowledge-intensive question answering.",
  "chunk_id": 19
 },
 {
  "doc_id": "citetrace",
  "title": "CiteTrace: Citation-Graph Retrieval-Augmented Generation for Scholarly Question Answering",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/citetrace",
  "topics": [
   "rag",
   "graph",
   "llm"
  ],
  "authors": [
   "Kian Roswell",
   "Mara Ellison"
  ],
  "text": "CiteTrace: Citation-Graph Retrieval-Augmented Generation for Scholarly Question Answering (SIGIR 2025). Authors: Kian Roswell, Mara Ellison. Answers scholarly research questions by retrieving over a citation graph: it follows citation links between papers to assemble a connected, well-grounded context for the language model, improving answer faithfulness and coverage for literature questions.",
  "chunk_id": 20
 },
 {
  "doc_id": "thriftroute",
  "title": "ThriftRoute: Cost-Aware Routing between Small and Large Language Models",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/thriftroute",
  "topics": [
   "llm",
   "efficiency"
  ],
  "authors": [
   "Tomas Beck",
   "Mara Ellison"
  ],
  "text": "ThriftRoute: Cost-Aware Routing between Small and Large Language Models (EMNLP 2025). Authors: Tomas Beck, Mara Ellison. Reduces expensive large-model API calls through selective invocation combined with online distillation, so a cheaper model handles more cases over time. ThriftRoute cuts serving cost substantially while preserving answer quality on routed workloads.",
  "chunk_id": 21
 },
 {
  "doc_id": "litmap",
  "title": "LitMap: Graph-Based Literature Discovery beyond Ranked Lists",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/litmap",
  "topics": [
   "graph",
   "retrieval"
  ],
  "authors": [
   "Ana Ferreira",
   "Mara Ellison"
  ],
  "text": "LitMap: Graph-Based Literature Discovery beyond Ranked Lists (CIKM 2024). Authors: Ana Ferreira, Mara Ellison. Treats literature discovery as navigating a graph over papers rather than returning a ranked list, letting researchers see the structure of a field and find related work by connections among papers, authors, and topics.",
  "chunk_id": 22
 },
 {
  "doc_id": "georeason",
  "title": "GeoReason: Spatial Retrieval-Augmented Generation for Real-World Spatial Questions",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/georeason",
  "topics": [
   "rag",
   "spatial",
   "llm"
  ],
  "authors": [
   "Priya Nandakumar",
   "Mara Ellison"
  ],
  "text": "GeoReason: Spatial Retrieval-Augmented Generation for Real-World Spatial Questions (ACL 2026). Authors: Priya Nandakumar, Mara Ellison. A retrieval-augmented generation framework that lets language models answer real-world spatial questions by retrieving and reasoning over geometric and geographic context rather than text alone.",
  "chunk_id": 23
 },
 {
  "doc_id": "tag-encoder",
  "title": "Text-Attributed Graph Encoders for Retrieval",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/tag-encoder",
  "topics": [
   "graph",
   "gnn"
  ],
  "authors": [
   "Ana Ferreira",
   "Sofia Marchetti",
   "Mara Ellison"
  ],
  "text": "Text-Attributed Graph Encoders for Retrieval (ICLR 2024). Authors: Ana Ferreira, Sofia Marchetti, Mara Ellison. Learns joint representations of node text and graph structure so text-attributed graphs can be encoded once and reused for retrieval and downstream reasoning, improving retrieval quality over text-only baselines.",
  "chunk_id": 24
 },
 {
  "doc_id": "gnn-explain",
  "title": "Faithful Explanations for Graph Neural Networks",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/gnn-explain",
  "topics": [
   "graph",
   "gnn",
   "explainability"
  ],
  "authors": [
   "Sofia Marchetti",
   "Mara Ellison"
  ],
  "text": "Faithful Explanations for Graph Neural Networks (NeurIPS 2024). Authors: Sofia Marchetti, Mara Ellison. Proposes an explanation method for graph neural networks that identifies the subgraph most responsible for a prediction while penalizing explanations the model does not actually rely on, improving faithfulness over saliency baselines.",
  "chunk_id": 25
 },
 {
  "doc_id": "online-distill",
  "title": "Online Distillation for Efficient LLM Serving",
  "source": "publication",
  "url": "https://www.example.edu/~mellison/online-distill",
  "topics": [
   "llm",
   "efficiency"
  ],
  "authors": [
   "Tomas Beck",
   "Kian Roswell",
   "Mara Ellison"
  ],
  "text": "Online Distillation for Efficient LLM Serving (MLSys 2026). Authors: Tomas Beck, Kian Roswell, Mara Ellison. Continuously distills a large model's behavior into a smaller student during deployment, so the cheaper model absorbs more of the workload over time without an offline retraining loop. Pairs naturally with cost-aware routing.",
  "chunk_id": 26
 }
]