FinanceBench evaluation of Mafin 2.5 (Powered by PageIndex)
-
Updated
Oct 20, 2025 - Python
#
financebench
Here are 14 public repositories matching this topic...
Advanced RAG pipelines for medical (HealthBench, MedCaseReasoning, MetaMedQA, PubMedQA) and financial (FinanceBench, Earnings Calls) QA. LangGraph orchestration + BAML structructed generation, Milvus Hybrid search (Dense + BM25 + RRF), three-layer Metadata Enrichment, Contextual AI instruction-following reranker, and DeepEval evaluation.
pubmed unstructured rag baml milvus earnings-calls contextual-ai llm langgraph rag-pipeline agentic-rag deepeval financebench healthbench
-
Updated
Sep 8, 2026 - Python
Multi-agent LangGraph RAG for financial Q&A — 72.7% on FinanceBench under κ=0.932 calibrated judge. RBAC at the vector layer, multi-party HITL on high-stakes answers, self-hosted LLM observability. pip install financebench-rag-agent
evaluation multi-agent-systems rag qdrant-vector-database retrieval-augmented-generation langfuse langgraph agentic-rag ragas-evaluation financebench
-
Updated
Jun 9, 2026 - Python
Finance-grade tool-calling agent over SEC filings for any US public company: exact XBRL numbers (never LLM arithmetic), every claim cited, abstains instead of fabricating. Validated on FinanceBench (44%→87% coverage, near-zero hallucination) with a 3-layer eval and a domain judge calibrated to human labels (κ=0.76). LangGraph + pgvector.
agent finance fintech xbrl sec-filings rag llm pgvector llm-evaluation langgraph tool-calling financebench
-
Updated
Aug 26, 2026 - Python
🔍 Empower efficient retrieval with PageIndex, a reasoning-based system that eliminates the need for vector databases and chunking for human-like results.
jquery agent pagination ai retrieval query-builder pager reasoning pageindex rag pageindexer qdrant llm vector-db agentic-ai financebench context-engineering zxb
-
Updated
Sep 10, 2026 - Jupyter Notebook
FinanceBench financial-report RAG with FAISS, Flask, Ragas, and financial numeric accuracy evaluation
-
Updated
Aug 30, 2026 - Python
Production RAG pipeline on FinanceBench: hybrid BM25+dense+RRF+Cohere Rerank, Anthropic Contextual Retrieval, conformal abstention (Yadkori 2024), Citations API grounding, triple eval (RAGAS+HHEM+DeepEval). Every eval number is reproducible and CI-gated.
-
Updated
Jun 16, 2026 - Python
a deliberately challenging benchmark built around real financial filings, where retrieving the right evidence is just as important as generating the answer.
-
Updated
Jul 27, 2026 - Jupyter Notebook
Grounded Q&A over financial filings (10-Ks/10-Qs): routes to the right document, decomposes the question, retrieves, and answers with page citations — or abstains when unsure. Simple-agentic RAG; FinanceBench hardest subset: 0.71 corpus / ~0.79 per-doc vs 0.19 naive.
python finance minimax rag qdrant llm document-qa retrieval-augmented-generation agentic-rag financebench
-
Updated
Jul 18, 2026 - Python
An end-to-end RAG pipeline that answers questions about SEC 10-K filings with grounded citations instead of hallucinated numbers. Built on FinanceBench: indexing with FAISS + BGE embeddings, generation with Llama-3.3-70B, three-axis evaluation (correctness / faithfulness / page-hit@k), improvement cycles.
-
Updated
Apr 29, 2026 - Jupyter Notebook
RAG pipeline for FinanceBench with retrieval, evaluation, improvement cycles, and chunk-size experiments.
-
Updated
Apr 27, 2026 - Jupyter Notebook
Rigorous evaluation of contextual retrieval techniques on FinanceBench: comparing 5 embedders × 4 chunking strategies with bootstrapped confidence intervals on FinMTEB and FinanceBench.
python benchmarking natural-language-processing information-retrieval pytorch embeddings semantic-search rag vector-search huggingface sentence-transformers retrieval-augmented-generation llm-evaluation contextual-retrieval late-chunking finance-nlp financebench finmteb
-
Updated
May 12, 2026 - Jupyter Notebook
Five RAG architectures built from scratch in plain Python — Naive, Hybrid, Agentic, Graph & production-grade Advanced — over real 10-K filings, evaluated on FinanceBench. No LangChain, no API keys, 100% local.
python embeddings knowledge-graph bm25 reranking rag vector-search hybrid-search qdrant llm local-llm document-qa retrieval-augmented-generation ollama semantic-chunking reciprocal-rank-fusion react-agent graph-rag agentic-rag financebench
-
Updated
Sep 9, 2026 - Python
A RAG pipeline built and evaluated on the FinanceBench dataset — retrieval, generation, and multi-dimensional evaluation across improvement cycles.
-
Updated
Aug 2, 2026 - Jupyter Notebook
Add this topic to your repo
To associate your repository with the financebench topic, visit your repo's landing page and select "manage topics."