AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Multimodal CoLRAG-TF: Triple-Filtered Retrieval for Complex PDFs

Takato Yasuno

Retrieval-augmented generation (RAG) over heterogeneous PDF collections remains challenging due to multimodal content, domain-specific terminology, and the need for multi-hop reasoning across dispersed evidence. We present Multimodal CoLRAG-TF, a four-axis fusion architecture that integrates dense text embeddings, BM25 keyword matching, knowledge-graph triple filtering, and image-based similarity for robust retrieval over complex documents. Our system constructs a multimodal index of 2,403 blocks extracted from 43 Japanese disaster lesson PDFs, supported by a hybrid OCR pipeline and LLM-based caption generation. To enhance compositional reasoning, we extract 11,414 OpenIE triples and index them with FAISS, enabling sub-second triple lookup and hierarchical propagation of relevance signals. A HippoRAG2-inspired coarse-to-fine retriever (volume $\to$ chapter $\to$ block) narrows the search space before final fusion scoring. Bayesian optimization over fusion weights reveals that the triple axis must dominate ($α_\text{triple} = 0.44$) to counteract lexical bias and sustain multi-hop retrieval quality. Evaluated on a 457-pair benchmark, Multimodal CoLRAG-TF achieves a Retrieval Recall of 0.9909 and a 71.6$\%$ improvement in multi-hop answer similarity over single-hop queries. An image-to-lesson pipeline using a vision LLM further demonstrates the applicability of the approach to visual inputs. These results show that triple-filtered multimodal fusion is essential for structured reasoning over noisy, heterogeneous PDFs and provides a general framework applicable beyond the disaster domain.

arXiv

Scaling Closed-Loop Feature Channel Configuration with LLMs

Tolgay Atinc Uzun, Radu Timofte, Dmitry Ignatov

Promising initial results in closed-loop large-language-model-based channel-configuration search demonstrated that neural-network widths can be optimized directly through executable code generation and accuracy feedback. However, those results were obtained from a relatively sparse set of valid evaluations, leaving open whether the observed optimization behavior transfers to a denser sampling regime and whether additional architectural regularities emerge when more generated networks are evaluated. To test this, the same search setting is scaled to 250 candidate networks per fine-tuning cycle. The analysis covers 2000 generated candidates from 8 complete cycles, yielding 462 verified CIFAR-100 evaluations after task and metadata filtering. Per-cycle mean accuracy exhibits a positive linear trend with slope 9.87e-4 (p=0.043), while the high-performing frontier improves more strongly: the best observed accuracy increases from 0.3144 to 0.3676, and both the top-5 and top-10 cycle-level means exhibit positive trends. The scaled run also reveals improved parameter efficiency. The best model reaches 0.3676 with 11.8M parameters, compared with an early high-performing model at 0.3144 with 166.5M parameters. Beyond accuracy, the larger sample exposes architectural regularities that were difficult to assess from sparse observations. Non-power-of-two channel widths occur in 41.8% of verified candidates, and the strongest models share structured channel-allocation patterns characterized by moderate early widths and expanded middle or later blocks. These findings indicate that the channel-search signal observed in the initial study transfers

ICML 2026

Reliability-Aware LLM Alignment from Inconsistent Human Feedback

Jingyi Huang, Ruohan Zong, Yujun Feng, Liran Ma, Lanyu Shang, Yang Zhang

Reinforcement Learning from Human Feedback (RLHF) is critical for aligning Large Language Models (LLMs) with human preferences. However, its efficacy is often compromised by the inherent inconsistency and subjectivity of human annotations. Existing preference optimization frameworks, such as Direct Preference Optimization (DPO), typically treat ambiguous pairs with high annotator disagreement identically to those with unanimous consensus, forcing models to overfit to inconsistent supervision signals and leading to suboptimal alignment. In this work, we propose $\textit{Reliability-Guided Preference Optimization}$ (RGPO), a robust framework designed to mitigate the impact of inconsistent human feedback. RGPO estimates annotator reliability and infers latent ground truth labels from noisy human feedback to identify robust preferences. Furthermore, we introduce a reliability-aware consistency optimization that dynamically modulates the training objective based on the consensus level of annotations, ensuring the model prioritizes high-consensus supervision signals. Extensive experiments on LLM alignment benchmarks demonstrate that RGPO effectively reduces inconsistency and noise in training data and achieves superior performance compared to widely adopted RLHF baselines. Our code and configurations are available at https://github.com/GenieHuang/RGPO.

arXiv

NMR Elucidation as an Agentic Search Problem, Not a Modeling Problem

Irina Espejo Morales, Damon Hinz, Marvin Alberts, Geraud Krawezik, Haewon Jeong, Shirley Ho

Structural elucidation from Nuclear Magnetic Resonance (NMR) data remains a fundamental bottleneck across chemistry, materials science, and biology. We demonstrate that an agentic AI system can perform this task at a level comparable to graduate-level chemistry students. Instead of training a model to directly map spectra to structures, we build a single autonomous agent, backed by a frozen LLM, that interacts with a curated environment with access to domain-specific processing tools, validation checks, tabulated chemical shifts, and instructions that outline the stepwise nature of a chemist's thinking process. On the Alberts dataset, our agent elucidates structures with a top-1 accuracy of 71%, comparable to the performance of graduate students at 66% top-1 accuracy. On the van Bramer and AstraZeneca datasets, our agent achieved 80% and 20% top-1 accuracy respectively, outperforming zero-shot end-to-end deep learning models which were trained on large datasets of simulated spectra. These results show that reframing NMR elucidation as an LLM-guided constrained search, rather than a modeling task, yields substantial gains and suggests a path toward multi-step orchestration frameworks that integrate a variety of tools, models, and domain knowledge to assist in automating spectroscopic analysis.

arXiv

Reproducing Recurrent Transformers: The CoTFormer

Aras Kavuncu, Bryan Vullo, Alberto Berni

The CoTFormer architecture formalizes Chain-of-Thought as a form of recurrent latent computation, preserving intermediate states as attendable representations to mimic explicit reasoning traces. In this work, we evaluate CoTFormer and its structural variants across perplexity and compute efficiency metrics. Furthermore, we extend evaluation to controlled algorithmic settings to determine whether this recurrent framework improves out-of-distribution generalisation on inductive reasoning tasks.

arXiv

ITPEval: Benchmarking Formal Translation Across Interactive Theorem Provers

Jiayi Wu, Robert Joseph George, Anima Anandkumar

Formal theorem proving has emerged as a frontier challenge for machine learning, yet the ecosystem is fragmented: proofs remain siloed across incompatible systems, limiting both training data for learning-based provers and the portability of verified results. We present ITPEval, the first benchmark for evaluating automated formal proof translation across four major ITPs (Lean 4, Rocq, Isabelle, and HOL Light), spanning two distinct logical foundations. Our benchmark comprises 1,560 source files and 6,848 theorems organized into a controlled tier of axiomatized files that isolates foundational translation difficulty, and an ecosystem tier drawn from real libraries that exposes API and proof-style mismatches. We release itpeval, a unified multi-ITP verification infrastructure with state-isolated warm backends that preserve per-artifact native checking semantics. We evaluate both statement and proof translation across five frontier and open-weight LLMs on 12 directed translation pairs: statement translation peaks at 29.1% pass@1 and proof translation at 10.5%; controlled theorems reach 29.7% proof pass@1 versus 5.2% for ecosystem-level translations, confirming that library mismatch is the dominant bottleneck. In addition to pass@k evaluation, a deterministic Lean 4 BEq check establishes equivalence for 54.0% of verified source-to-Lean 4 miniF2F statement translations, showing that native type-checking alone can substantially overestimate semantic fidelity; in an autoformalization/auto-informalization round-trip study, Rocq and HOL Light are easier formalization targets than Lean 4 and Isabelle, while multi-ITP context improves pooled Lean 4 success from 4.8% to 10.6%. Our benchmark, verification infrastructure, and evaluation pipelines are publicly released.

arXiv

Structured Latent Space Modeling over Multi-Scale Temporal Patches for Multivariate Time Series Forecasting

Xingsheng Chen, Deyu Yi, Siu-Ming Yiu

Multivariate time series encode structural patterns that unfold across multiple temporal scales, yet most forecasting backbones treat learned representations as transient byproducts of prediction, leaving the organizational geometry of these patterns underexploited. We introduce M2Patch, a CNN-based forecasting architecture that maps channel-independent multivariate observations into a structured latent space through two complementary differentiable constraints. Multi-scale patching decomposes the input into overlapping temporal granularities; depthwise separable convolutions with progressive dilation extract scale-specific features in linear time; and per-scale learned projections compress these features into a compact latent representation. The latent space is organized by an intra-scale smoothness constraint that enforces temporal continuity between adjacent patches, and an inter-scale alignment constraint, realized through learnable cross-scale mappings, that restores cross-granularity interaction within the channel-independent design, ensuring that all scales encode mutually consistent representations of the underlying dynamics. Experiments on ten real-world benchmarks show that M2Patch achieves 57 best and 34 second-best results across 40 forecasting settings, matching or exceeding representative baselines on most benchmarks while maintaining linear computational complexity and robustness to patch-level input corruption.

arXiv

Recovering Clinical Utility Under Differential Privacy: Empirical Validation of Adaptive Federated Aggregation on Heterogeneous Cardiovascular Datasets

Rodrigo Tertulino, Laercio Alencar, Ricardo Almeida

Validating federated learning frameworks on real clinical data is an essential step between proof-of-concept demonstrations in controlled synthetic environments and deployment in real multicenter healthcare settings. A prior architectural study by the same authors (Tertulino and Alencar, 2026) demonstrated, on a synthetic six-feature benchmark, that server-side adaptive optimization acts as a temporal denoiser for Differential Privacy noise, answering an open challenge identified in the original pipeline work (Tertulino, 2025). That study used synthetically generated data and explicitly identified real-world validation as a priority future direction. The present work addresses this gap by validating the FedCVR framework on five publicly available real cardiovascular datasets (Framingham, Cleveland, Hungarian, Switzerland, and Long Beach VA), harmonized to the 13-attribute UCI Heart Disease schema and configured as a heterogeneous federated scenario with leave-one-institution-out cross-validation. Results demonstrate that FedCVR preserves its adaptive advantage on real data, achieving an F1-Score of 79.2% and AUC of 0.96 under the operational privacy budget (noise multiplier = 0.8, privacy budget epsilon approximately 4.2), while statistically outperforming standard FedAvg on all evaluated metrics (paired t-tests, all p <= 0.003, significant under the Bonferroni-corrected threshold). The measured privacy cost on real data confirms the graceful degradation pattern observed in the synthetic experiments, providing empirical evidence of the framework's clinical viability in genuine multicenter contexts.

arXiv

Physics-Informed Feature Engineering 1D-CNN for Multilayer Cloud Detection from Geostationary Satellites

Fu Wang, Chi Yang, Qi-Feng Lu, Rui-Xia Liu, Xiao-Fei Yang, Xiao-Fang Liu, Bo Li, Lin Chen

Multilayer cloud detection from active--passive observation is vital for numerical weather prediction. In this study, channel selections derived from threshold-based algorithms are embedded as feature-engineering priors into a 1D-CNN, and machine learning (ML) is used to learn latent physical relationships to simplify physical retrievals for operational deployment. The results show that the 1D-CNN achieves a multilayer-cloud probability of detection ($\mathrm{POD}{\mathrm{mul}}$) of 0.620 and a false alarm rate ($\mathrm{FAR}{\mathrm{mul}}$) of 0.240, outperforming the conventional threshold algorithm ($\mathrm{POD}{\mathrm{mul}} = 0.558$, $\mathrm{FAR}{\mathrm{mul}} = 0.369$). These results demonstrate that prior physical knowledge derived from radiative transfer theory can serve as an effective feature-engineering prior. Further experiments show that ML-revealed physical mechanisms can also enhance traditional algorithms. Replacing AGRI channel 12 (C12, centered at $10.8~μ\mathrm{m}$) with channel 13 (C13, centered at $12.0~μ\mathrm{m}$) increased $\mathrm{POD}{\mathrm{mul}}$ from 0.558 to 0.609 without materially affecting $\mathrm{FAR}{\mathrm{mul}}$. However, for AHI, substituting the $11.2~μ\mathrm{m}$ channel with the $12.3~μ\mathrm{m}$ channel yielded negligible improvement. In addition to spectral response function (SRF) mismatches, a primary contributing factor is the channels' on-orbit radiometric stability. Hence, physics-informed machine-learning methods appear promising for advancing remote-sensing AI, while sensor-specific characteristics must be considered during operational transfer.

arXiv

From Intent to Infrastructure: LLM-Driven Agent Compilers for ISAC Networks

Lijie Zheng, Xudong Zhong, Baoquan Ren, Xiangwu Gong, Xinghui Zhu, Ji He

Integrated sensing and communications (ISAC) is moving from proof-of-concept demonstrations to system-level deployment in sixth-generation (6G) networks. Because sensing and communication share hardware, spectrum, and waveform resources, ISAC design now involves many tightly coupled choices, including waveform selection, sensing algorithm setup, resource scheduling, and deployment planning. This design space is already too large to manage well through manual tuning or isolated optimizers. This article introduces the \textit{Agent Compiler}, a large language model (LLM)-enabled compilation layer that translates high-level engineering intent into complete and executable ISAC system configurations. The Agent Compiler works in four stages: intent parsing, task decomposition, policy graph synthesis, and infrastructure mapping. It produces a verifiable intermediate representation called the ISAC Policy Graph (IPG). A runtime engine then deploys the compiled configuration and supports closed-loop adaptation at three levels: fast parameter updates, partial recompilation of affected subgraphs, and full workflow recompilation. The core design principle is strict time-scale separation: the LLM handles slow-loop strategic decisions, while proven algorithms retain real-time control in the fast loop. A UAV-assisted disaster rescue example illustrates the full compilation process. We also discuss open issues, including compilation latency, output reliability, constraint verification, and pipeline security, to guide future research.

arXiv

Assessing the Operational Impact of Poisoning Attacks over Augmented 3D Point Cloud Public Datasets for Connected and Autonomous Vehicles

Marwan Lazrag, Badis Hammi, Lorena Gonzalez-Manzano, Joaquin Garcia-Alfaro

Poisoning attacks against public datasets lead to major concerns, such as (i) misclassification of perceived objects when the poisoned data is used for training and (ii) embedding of backdoors that may eventually be triggered later on, when specific conditions in the system apply over the learned models. Its impact over data augmentation models is unclear. While data augmentation reduces the likelihood of poisoning attack success, some valid questions remain. Is data augmentation affecting the impact of poisoning attacks? can it increase the number of poisoned samples or injected backdoors? We explore in this paper some of these questions. We assess the effects of augmenting poisoned 3D point cloud datasets and validate that poisoning is able to evade the sanitizing nature of augmentation techniques when using the concrete case of Generative Adversarial Network (GAN) techniques to exemplify the case of data augmentation processing. We also validate that poisoning propagates over the augmented datasets and perturbs the decision made by general-purpose classifiers, in the end. All the experimental material (including tools, datasets, and classifiers) is publicly available, to facilitate reproducibility and to foster further research in the topic.

arXiv

MLLM-DataEngine: Closing the Loop of Multimodal Instruction Tuning Data Generation

Zhiyuan Zhao, Bin Wang, Linke Ouyang, Yiqi Lin, Pan Zhang, Xiaoyi Dong, Jiaqi Wang, Conghui He

In this paper, we propose MLLM-DataEngine, a novel closed-loop system that bridges data generation, model training, and evaluation. Within each loop iteration, the MLLM-DataEngine first analyzes the weakness of the model based on the evaluation results, then generates a proper incremental dataset for the next training iteration, and enhances the model capability iteratively. Compared with previous instruction fine-tuning dataset collection methods which are separate from the benchmarking, MLLM-DataEngine shows better targeting and can improve MLLMs's capabilities more effectively. Firstly, we propose an Adaptive Bad-case Sampling module, which can effectively analyze model weakness based on the benchmarking results and adjust the generation of incremental datasets flexibly. Secondly, in order to ensure high-quality data for specific capability types, the most representative in-context examples and abundant information are provided to GPT-4, which helps GPT-4 fully comprehend the model's weakness and further guarantees high-quality generated data. Through extensive experiments, we find MLLM-DataEngine could boost the MLLMs capability in a targeted and automatic manner without human participants. We hope MLLM-DataEngine could be a general solution for the following MLLMs data curation. Code, data, and model are available at https://github.com/opendatalab/MLLM-DataEngine.

arXiv

Scaling Quantum Machine Learning without Tricks: Full-Resolution and Diverse Image Generation

Jonas Jäger, Florian J. Kiwit, Carlos A. Riofrío

Quantum generative modeling is a rapidly evolving discipline at the intersection of quantum computing and machine learning. Contemporary quantum machine learning is generally limited to toy examples or heavily restricted datasets with few elements. This is not only due to the current limitations of available quantum hardware but also due to the absence of inductive biases arising from application-agnostic designs. Current quantum solutions must resort to tricks to scale down high-resolution images, such as relying heavily on dimensionality reduction or utilizing multiple quantum models for low-resolution image patches. Building on recent developments in classical image loading to quantum computers, we circumvent these limitations and train quantum Wasserstein GANs on the established classical MNIST and Fashion-MNIST datasets. Using the complete datasets, our system generates full-resolution images across all ten classes and establishes a new state-of-the-art performance with a single end-to-end quantum generator without tricks. As a proof-of-principle, we also demonstrate that our approach can be extended to color images, exemplified on the Street View House Numbers dataset. We analyze how the choice of variational circuit architecture introduces inductive biases, which crucially unlock this performance. Furthermore, enhanced noise input techniques enable highly diverse image generation while maintaining quality. Finally, we show promising results even under quantum shot noise conditions.

arXiv

RL-ADA: A World-Feedback Framework for Adversarially Robust Enterprise Dialogue Agents

Ram Narayanan, Harshit Rajgarhia, Abhishek Mukherji

Deploying task-oriented dialogue agents in enterprise customer support faces a persistent annotation bottleneck: robust training requires labelled interaction data at scale, yet enterprise conversational logs are privacy-sensitive and expensive to annotate, while user behaviour evolves faster than labelling pipelines can keep pace. We present RL-ADA (Reinforcement Learning with Adversarial Dialogue Agents), a co-evolutionary training framework that eliminates this bottleneck by replacing human labels with \emph{world feedback}: consequence-based reward signals derived directly from measurable interaction outcomes. A Customer Support Agent (DA, 3B parameters) and an Adversarial Customer Agent (CA, 7B parameters) co-evolve in an adversarial arena guided by a fixed automated judge: the DA is rewarded for correctly handling multi-turn customer conversations to successful resolution, while the CA is rewarded for producing realistic, intent-concealing utterances that cause misroutes, creating asymmetric adversarial pressure through opposing but independently structured rewards. An isolation gym iteratively retrains the weaker agent on prior-failure transcripts, requiring no human annotation at any stage. In a banking customer support proof of concept, tool-routing errors are eliminated and the strict end-to-end PASS rate doubles over five co-evolutionary cycles, driven solely by automated arena reward with no labelled data. We additionally observe the emergence of \textbf{Contextual Camouflage}, an adversarial strategy in which the CA learns to embed intent within dense realistic customer detail purely from reward pressure, with direct implications for enterprise red-teaming and robustness evaluation.

arXiv

Dual-Form ASR: Semantics-Aware Inverse Text Normalization for Chinese Speech Recognition

Fengrun Zhang, Li Fu, Wangjin Zhou, Lu Fan, Youzheng Wu, Xiaodong He

Modern automatic speech recognition (ASR) scenarios require both spoken-form transcripts for faithful transcription and readable written-form transcripts with inverse text normalization (ITN). However, these forms are typically produced by cascaded modules, where a spoken-form ASR output is rewritten by a separate ITN component, making written-form ASR-ITN vulnerable to recognition errors and decoupling normalization from acoustic-contextual modeling, especially for semantically dependent numeric expressions. In this paper, we propose Dual-Form ASR (DF-ASR), a framework that extends spoken-form ASR capability to semantics-aware written-form ITN through paired spoken-form and written-form supervision while retaining prompt-level selection between transcript forms. The dual-form supervision is constructed via a large language model (LLM)-driven generate-and-judge workflow, and training is further enhanced by ITN-MWER, a sequence-level objective that assigns higher cost to errors on normalization-sensitive spans. We also introduce a decision-aware REQUIRE-ITN/\FORBID-ITN protocol to separately measure required normalization and forbidden-span preservation. On manually annotated Chinese subsets from SpeechIO, DF-ASR consistently outperforms open-source ASR-ITN systems, remains competitive with strong closed-source references, and preserves reliable prompt-level control between spoken-form and written-form outputs.

arXiv

AI Evaluation Should Work With Humans

Jan Kulveit, Gavin Leech, Tomáš Gavenčiak, Raymond Douglas

This position paper argues that the dominant paradigm of AI evaluation (which focuses on superhuman autonomous performance and so implicitly targets the goal of replacing humans) is guiding AI development in the wrong direction. Instead, the AI community should pivot to evaluating the performance of human--AI teams. We argue that this collaborative shift will foster AI systems that act as true complements to human capabilities and therefore lead to far better societal outcomes than will the current process.

arXiv

BCIJelly: An integrated ecosystem for brain-computer interface research

Liyuan Han, Xinrui Yang, Tianyu Zheng, Qizhi Yang, Yitao Qin, Liang Chen, Qinglai Wei, Binjie Hong, Xinhe Zhang, Rui Xiong, Yong Gu, Mu-ming Poo, bo xu, Chengyu Li, Tielin Zhang

Brain-computer interface (BCI) research relies on multistage computational pipelines, yet progress remains constrained by fragmented data formats, heterogeneous decoder implementations and hardware-specific deployment toolchains, and researchers lack an integrated workflow. Here, we fill this gap with BCIJelly, a unified computational ecosystem that integrates 18 curated BCI datasets, 15 benchmark decoders and an algorithmic library of 80 reusable modules, an automated architecture search (AAS) procedure, and hardware-aware deployment through the toChip pipeline within a single Python framework. AAS constructs task-specific decoders without manual architecture design. It is further extended into a closed-loop mode guided by a large language model (LLM), which uses task specifications, module descriptions and search history to support multitask and cross-species decoding. The toChip pipeline compiles trained decoders for execution on neuromorphic chips, enabling energy-efficient deployment for BCI systems. An accompanying visualization software provides a graphical interface to the full workflow, making BCIJelly accessible without programming. We validate BCIJelly across five BCI paradigms (motor, visual, speech, emotion and auditory) with recordings from humans, macaques and mice, and single-task, multitask and cross-species decoding settings. BCIJelly establishes a unified and extensible infrastructure that bridges decoder development and hardware-aware deployment for BCI research.

arXiv

EU-ETS under attack? The impact of carbon price suppression on the decarbonization of the power sector

Javier Gonzalez-Ruiz, Carlos Rodriguez-Pardo, Alice Di Bella, Paolo Mastropietro, Jose Pablo Chavez-Avila, Massimo Tavoni

European countries are debating policies to mitigate the increased energy costs caused by renewed geopolitical tensions, while pursuing decarbonization and electrification. A notable example is Italy's 2026 Decreto Bollette package, which proposes to remove the carbon price equivalent from the bids of certain gas-driven power plants to wholesale electricity markets, among other provisions. We use this as a case study to assess the long-term implications of suppressing the carbon price signal in the electricity market for investment, emissions, and consumer costs. We employ a stylized Italian power system using MARLEY, a multi-agent reinforcement learning framework focused on long-term electricity market assessments. In this framework, we test this policy across configurations with varying levels of support for green investment, resource adequacy, and flexibility. Results show that partial suppression of the carbon price signal yields short-term cost reductions but only a minor long-term effect on total system costs, as the deferred emissions are ultimately repaid by consumers. CO$_2$ emissions rise across most configurations since suppressing the price signal erodes incentives for renewable and storage investment. Only the most ambitious configurations for supporting green investment avoid this outcome, but they do so by marginalizing the wholesale price signal itself, thereby requiring a commitment to a hybrid market paradigm that is in contradiction with the rationale of the proposed price intervention.

arXiv

KumbhDoot: A Scale-Ready, LLM-Bounded Architecture for Mass-Gathering Public-Service Assistants

Saurabh Sakalkar, Abhishek Singh, Ramesh Raskar

Mass religious gatherings such as the Kumbh Mela concentrate tens of millions of people into a single region over a few weeks, producing intense, repetitive, multilingual, and safety-critical demand for information. The default response, a conversational assistant that routes every query to a large language model (LLM), is poorly matched to this setting: it is costly at scale, slow on emergency paths, prone to hallucination on facts that can cause physical harm, and unusable when connectivity fails. We describe KumbhDoot, an agentic pilgrim assistant for the Nashik Simhastha Kumbh Mela built on a different principle. It operates on a foundational design principle that prioritizes semantic similarity over starting with an LLM. Generative models are invoked only in instances where similarity-based retrieval is insufficient to produce a correct answer. The system utilizes a "semantic cache": an embedding-indexed store as a single retrieval primitive, which handles intent routing, answer caching, offline lookups, and multi-agent retrieval. A custom three-tier agent architecture operates directly on this store, ensuring decision paths remain inspectable and avoiding the use of generic multi-agent frameworks that would trigger implicit per-step LLM calls. We present the architecture, an analytical cost model for its per-query economics, and an honest account of where similarity is sufficient and where generative reasoning remains necessary. We argue that for bounded, high-stakes, low-connectivity public-service domains, a similarity-first and LLM-bounded design is not merely cheaper but architecturally more appropriate than an LLM-default one.

arXiv

MAM-AI: An On-Device Medical Retrieval-Augmented Generation System for Nurses and Midwives in Zanzibar

Yi Ren

Maternal and newborn mortality remain among the highest in sub-Saharan Africa, where midwifery care is often delivered by nurses who lack midwifery training to international standards, and consulting authoritative guidance at the point of care is hard: the guidelines are long and connectivity is intermittent. We present MAM-AI, a medical question-answering assistant for nurse-midwives in Zanzibar that runs entirely on a commodity Android device: a question is embedded (EmbeddingGemma, 300M) and matched against a curated corpus of 87 guideline documents (63,650 passages), then answered with citations by a 4B int4 generator (Gemma 4 E4B), fully offline, with no query leaving the device. We evaluate the exact deployed configuration with a layered methodology -- retriever, generator under oracle context, end-to-end, and latency -- scored by LLM judges validated against physician rubrics. The evaluation relocates the hard problem. On-device retrieval is essentially solved: the 300M embedder ranks third of seven retrievers and rivals cloud systems, so the passages the system needs are usually found. The small generator is what remains in doubt: adding retrieved context does not improve its answers, and at 4B it cannot be both helpful and safe at once -- of two same-size candidates, the more helpful one commits genuine dangerous errors, so we deploy the other, which is about twice as faithful to its sources (as faithful as a frontier model), and recover its helpfulness with a redesigned prompt that cuts deflection from 33% to 3%. Corpus quality is decisive for the same reason: where the corpus holds the right passage the answer is specific and actionable, and where it does not it goes vague. MAM-AI is a thoroughly evaluated, open-source research prototype, not a fielded product; the system, knowledge base, benchmarks, and evaluation harness are released.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.