Grounded Graph RAG · Research · Scenarios · Deals
BankingAI ingests credit agreements, equity research, and filings into a hierarchical section graph — then answers with hybrid retrieval, citation markers, confidence scores, and a hallucination fallback so analysts can trust what they ship into deals and scenarios.
At scale, document retrieval quality beats a frontier model alone. BankingAI layers ingest, two-stage retrieval, confidence, constrained generation, citations, and continuous evals so answers stay auditable for credit, equity, and deal workflows.
PDF/DOCX/MD → hierarchical sections with profile-aware summaries, content hashes, and graph edges.
BM25 keywords catch covenant language; dense embeddings catch paraphrase — fused per query.
Stage-1 ANN candidates + stage-2 cross-encoder-style rerank so the right section floats to the top.
Calibrated confidence on each hit and answer-level scores that gate low-evidence responses.
Schema-guided JSON answers so the model must return structured claims, not free-form drift.
Inline [n] markers bound to section IDs, quotes, and the documents your team actually ingested.
If support or confidence is weak, BankingAI abstains or rewrites instead of inventing numbers.
Hit@k / MRR / grounding checks on golden banking queries — quality is measurable, not hoped for.
Section KV cache, query-result cache, and chat session memory for faster multi-turn analysis.
Trace IDs and stage timings from retrieve → rerank → generate so you can debug trust failures.
Outcome for analysts: faster memo Q&A with provenance you can defend in IC — fewer fabricated covenants, multiples, or DSCR claims.
One ingestion pipeline, three tailored lenses on the same source material.
Surface covenants, leverage grids, syndication terms, and closing conditions without re-reading every amendment. Escrow milestones to the sections that matter at close.
Map valuation drivers, guidance, and catalysts across filings and models. Hybrid search finds revenue language even when terminology shifts quarter to quarter.
Track capital ratios, NPL migration, and regulatory exposure in credit agreements and bank supplements — with KV-cache reuse when documents update incrementally.
BankingAI combines grounded Graph RAG, analyst workflow UX, scenario quant, and blockchain execution in one stack.
Hierarchical sections with analyst-tuned summaries and metadata fields.
Skip re-inference on unchanged sections; chain prefixes for long documents.
Hybrid BM25 + embeddings, ANN candidate search, rerank, citations, and abstain gates.
Anchor Solana devnet contracts to content hashes with milestone schedules.
Designed for investment banking, credit, and equity workflows from thesis ideation through execution.
Analyst login, thesis input, macro regime selection, and AI-suggested starter scenarios with assumptions prefilled.
Run profile-aware market research and attach citations, then transfer the thesis directly into scenario modeling.
Request scenario analysis in chat and view DSCR / IRR visuals inline before opening deeper workspaces.
Backtests, peer comps, downside/upside cases, alternatives, and Monte Carlo distributions with sensitivity analysis.
Turn analysis into coordinated actions across counterparties, contracts, and payment gates.
Deal originator, counterparty, and executor agents hand off tasks with auditable JSON task payloads.
Expose ingest, search, and contract tools to Cursor and other MCP clients for operator workflows.
Gate paid execution endpoints using x402-compatible payment headers before contract settlement.
Preview escrow PDA and program references for transparent, verifiable on-chain execution context.
Start with mock mode for fast demos, then scale to hosted LLM and production infrastructure.
Run full product walkthroughs without GPU requirements, ideal for pilots and client previews.
Switch to self-hosted model backends for richer response quality and longer-context analysis.
Serve interactive app/backend on Fly while mirroring marketing pages via GitHub Pages.
Capture signups with backend persistence and trigger welcome-email workflows when configured.
Chat + Scenario Builder + quant modeling + A2A + Solana escrow + Solscan proof in one walkthrough.
The latest demo includes the new Scenario Builder experience with thesis research, AI scenario suggestions, and chart-driven modeling.
Join the waitlist for hosted workspaces, team profiles, and production Solana mainnet escrows.
Google sign-in loads when configured.