papers/hanzo-search/sections/99-conclusion.tex

\section{Conclusion}

\label{sec:conclusion}

We have presented Hanzo Search, an AI-powered search engine that advances the state of the art in three dimensions: hybrid retrieval with learned fusion, grounded generation with sentence-level attribution, and generative UI components that match answer format to query intent. Our evaluation demonstrates improvements of 18.7\% in NDCG@10 over single-method retrieval, 94.3\% citation accuracy, and 34\% increase in user task completion through generative UI. Production deployment over 12 months with 2.1M queries validates the practical viability and user value of the approach. Hanzo Search is available at \texttt{search.hanzo.ai}.

\bibliographystyle{plain}
\begin{thebibliography}{30}

\bibitem{asai2023self}
A.~Asai, Z.~Wu, Y.~Wang, A.~Sil, and H.~Hajishirzi.
\newblock Self-{RAG}: Learning to retrieve, generate, and critique through self-reflection.
\newblock In \emph{ICLR}, 2024.

\bibitem{cormack2009reciprocal}
G.~V. Cormack, C.~L.~A. Clarke, and S.~B\"uttcher.
\newblock Reciprocal rank fusion outperforms condorcet and individual rank learning methods.
\newblock In \emph{SIGIR}, 2009.

\bibitem{formal2022splade}
T.~Formal, C.~Lassance, B.~Piwowarski, and S.~Clinchant.
\newblock {SPLADE} v2: Sparse lexical and expansion model for information retrieval.
\newblock In \emph{SIGIR}, 2022.

\bibitem{gao2023alce}
T.~Gao, H.~Yen, J.~Yu, and D.~Chen.
\newblock Enabling large language models to generate text with citations.
\newblock In \emph{EMNLP}, 2023.

\bibitem{gao2023rarr}
L.~Gao, Z.~Dai, P.~Pasupat, et~al.
\newblock {RARR}: Researching and revising what language models say, using language models.
\newblock In \emph{ACL}, 2023.

\bibitem{google2023sge}
Google.
\newblock Search generative experience.
\newblock \emph{Google Blog}, 2023.

\bibitem{guu2020realm}
K.~Guu, K.~Lee, Z.~Tung, P.~Pasupat, and M.~Chang.
\newblock {REALM}: Retrieval-augmented language model pre-training.
\newblock In \emph{ICML}, 2020.

\bibitem{he2021debertav3}
P.~He, J.~Gao, and W.~Chen.
\newblock {DeBERTaV3}: Improving {DeBERTa} using {ELECTRA}-style pre-training with gradient-disentangled embedding sharing.
\newblock \emph{arXiv preprint arXiv:2111.09543}, 2021.

\bibitem{huang2023hallucination}
L.~Huang, W.~Yu, W.~Ma, et~al.
\newblock A survey on hallucination in large language models.
\newblock \emph{arXiv preprint arXiv:2311.05232}, 2023.

\bibitem{izacard2023atlas}
G.~Izacard, P.~Lewis, M.~Lomeli, et~al.
\newblock Atlas: Few-shot learning with retrieval augmented language models.
\newblock \emph{JMLR}, 24(251):1--43, 2023.

\bibitem{kwiatkowski2019natural}
T.~Kwiatkowski, J.~Palomaki, O.~Redfield, et~al.
\newblock Natural questions: A benchmark for question answering research.
\newblock \emph{TACL}, 7:453--466, 2019.

\bibitem{lewis2020retrieval}
P.~Lewis, E.~Perez, A.~Piktus, et~al.
\newblock Retrieval-augmented generation for knowledge-intensive {NLP} tasks.
\newblock In \emph{NeurIPS}, 2020.

\bibitem{malkov2018efficient}
Y.~A. Malkov and D.~A. Yashunin.
\newblock Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs.
\newblock \emph{IEEE TPAMI}, 42(4):824--836, 2018.

\bibitem{metzler2021rethinking}
D.~Metzler, Y.~Tay, D.~Bahri, and M.~Najork.
\newblock Rethinking search: Making domain experts out of dilettantes.
\newblock \emph{SIGIR Forum}, 55(1), 2021.

\bibitem{nair2023bing}
V.~Nair and S.~Mehdi.
\newblock Reinventing search with a new {AI}-powered {Bing} and {Edge}.
\newblock \emph{Microsoft Blog}, 2023.

\bibitem{nakano2021webgpt}
R.~Nakano, J.~Hilton, S.~Balaji, et~al.
\newblock {WebGPT}: Browser-assisted question-answering with human feedback.
\newblock \emph{arXiv preprint arXiv:2112.09332}, 2021.

\bibitem{perplexity2023}
Perplexity~AI.
\newblock Perplexity: Ask anything.
\newblock \emph{Perplexity Documentation}, 2023.

\bibitem{santhanam2022colbertv2}
K.~Santhanam, O.~Khattab, J.~Saad-Falcon, et~al.
\newblock {ColBERTv2}: Effective and efficient retrieval via lightweight late interaction.
\newblock In \emph{NAACL}, 2022.

\bibitem{shi2023replug}
W.~Shi, S.~Min, M.~Yasunaga, et~al.
\newblock {REPLUG}: Retrieval-augmented black-box language models.
\newblock \emph{arXiv preprint arXiv:2301.12652}, 2023.

\bibitem{thakur2021beir}
N.~Thakur, N.~Reimers, A.~R\"uckl\'e, A.~Srivastava, and I.~Gurevych.
\newblock {BEIR}: A heterogeneous benchmark for zero-shot evaluation of information retrieval models.
\newblock In \emph{NeurIPS Datasets and Benchmarks}, 2021.

\bibitem{xiao2023bge}
S.~Xiao, Z.~Liu, P.~Zhang, and N.~Muennighoff.
\newblock {C-Pack}: Packaged resources to advance general {Chinese} embedding.
\newblock \emph{arXiv preprint arXiv:2309.07597}, 2023.

\bibitem{you2023}
You.com.
\newblock You.com: The {AI} search engine you control.
\newblock \emph{You.com Documentation}, 2023.

\bibitem{robertson2009bm25}
S.~Robertson and H.~Zaragoza.
\newblock The probabilistic relevance framework: {BM25} and beyond.
\newblock \emph{Foundations and Trends in Information Retrieval}, 3(4):333--389, 2009.

\bibitem{nogueira2020document}
R.~Nogueira, Z.~Jiang, R.~Pradeep, and J.~Lin.
\newblock Document ranking with a pretrained sequence-to-sequence model.
\newblock In \emph{EMNLP Findings}, 2020.

\bibitem{karpukhin2020dense}
V.~Karpukhin, B.~Oguz, S.~Min, et~al.
\newblock Dense passage retrieval for open-domain question answering.
\newblock In \emph{EMNLP}, 2020.

\end{thebibliography}