Research agentsfor finance.

YSquared builds an agent workforce that reads, verifies, and cites financial documents with the rigor of a trained analyst — so your team spends its time on judgment, not reading.

research_brief.md
>What is NVIDIA's data-center
revenue concentration risk?
─────────────────────────────
·NVDA_10K_FY2024.pdf
·NVDA_Q3_2024_Transcript.pdf
·SEC 13F Filings (Q3 2024)
>Agent: DeepResearch v2.1

A system built forevidence-grade research.

01

Evidence Pipeline

Documents never go straight to a model. Filings, transcripts, and data exports are parsed into a structured evidence layer — tables, entities, segments — that every downstream step queries and cites.

> pipeline ingest NVDA_10K_2024.pdf
  parse      ✓ 847 segments
  normalize  ✓ 23 tables
  index      ✓ evidence graph built
02

Verification Gates

Every figure clears a gate before it reaches a draft: cross-checked against its source table, reconciled across documents. Discrepancies are flagged, never silently resolved.

> gate check draft_003
  $47,532M ← 10-K Item 7  ✓ pass
  $47.5B (rounded)        ✗ blocked
  1 claim returned for evidence
03

Provenance by Construction

Citations aren't added after the fact — findings can only be assembled from evidence that carries its source. Every claim traces to the exact paragraph, table row, or footnote.

> trace finding_001
  ← 10-K p.47 ¶2 · conf 0.97
  ← Transcript Q2:14 · conf 0.91
04

Adaptive Workflows

Research plans aren't fixed scripts. Workflows branch on what the evidence shows — fanning out readers in parallel, re-running weak steps, escalating when sources conflict.

> run brief_001.yaml
  fan-out    3 readers   ✓ 2.1s
  conflict   guidance    → re-read
  escalate   1 item      → analyst

From briefto findings.

01

Ingest

Upload or connect documents: 10-Ks, transcripts, expert calls, data exports. The agent builds a structured knowledge graph per issuer.

02

Brief

Define your research question in plain English. The agent drafts an execution plan and confirms scope before proceeding.

03

Research

Agents read, compare, and synthesize across sources simultaneously. Watch the process in real time or check back when complete.

04

Validate

Every figure is cross-referenced to its source. Discrepancies are flagged, not silently resolved. Verification gates run on every claim automatically.

05

Deliver

Receive a structured research brief with inline citations and an audit trail. Export to PDF, Markdown, or your internal workflow.

Watch an agentwork a brief.

A replay of an agent run, end to end — plan, evidence, findings. Every figure traced to its source.

nvda_concentration.run
>How concentrated is NVIDIA's data-center revenue — and how exposed is it to a single hyperscaler pulling back?
·NVDA_10K_FY2024.pdf
·NVDA_Q3_FY2024_Transcript.pdf
·SEC_13F_Q3_2024/

Built for thefinance research workflow.

Equity Research

Automate the reading workload for coverage initiation and earnings updates. Generate segment analysis, comps tables, and management tone summaries.

NVDA · Q3 FY2024 Analysis
Revenue beat: +$2.1B vs consensus
GM expansion: +310bps q/q

Credit & Private Credit

Extract covenant packages, EBITDA build-ups, and leverage progression across deal documents. Flag definition drift between vintages.

Covenant tracker · TLB 2024-A
≤ 7.5x Net Leverage  ✓ 5.8x
Minimum Liquidity $50M  ✓ $87M

M&A Diligence

Read data room documents at scale. Surface reps-and-warranties risk, customer concentration, and revenue quality issues across hundreds of files.

Diligence scope: 340 files
Customer conc. flag: Cust A = 34%
Revenue quality: 82% recurring

Macro & Policy

Monitor central bank communications, regulatory filings, and legislative text. Track semantic shifts in guidance language over time.

Fed Minutes drift analysis
“transitory” mentions: ↓ 94%
Rate path language: hawkish +2.1σ

Finance research isdrowning in reading.

The average equity analyst spends more than half their time reading—not thinking. Earnings transcripts, 10-K risk factors, proxy statements, expert call summaries. The signal is there. Finding it is the problem.

YSquared Research was built to solve exactly this. Our agents don’t summarize—they read with the rigor of a trained analyst: tracking figures across documents, catching restatements, sourcing every claim.

We are a small technical team, founded in 2026, building toward design partnerships with research teams at hedge funds, asset managers, and banks — agents calibrated to your workflow, your documents, and your definition of quality.

The reading is the easy part—for an agent.

Talk to the team.

We work with a small number of research teams at a time. If your team spends too much time reading and not enough time deciding, we’d like to hear from you.

Investors and cloud partners: same calendar — we're happy to walk through the platform and roadmap.