Meridian Capital Partners is a US mid-market investment firm whose analysts own the full modeling stack — from raw SEC EDGAR data ingestion through investment committee outputs — without a dedicated data-engineering layer. David Park covers equity research and M&A support, building integrated 3-statement financial models from public company filings as part of every pre-investment review. Speed and accuracy at the pre-modeling stage directly determine how many companies the team can evaluate within a given deal window.
XBRL tag proliferation broke every generic template
The raw EDGAR company-facts JSON for a single company contains dozens of overlapping tags for the same line item — multiple representations of net income, several depreciation-and-amortization variants, inconsistent tag hierarchies across reporting periods. Selecting the wrong tag compounds silently through the model until the balance sheet fails to close.
Beyond tag selection, inter-statement connections are easy to get wrong: net income must flow from the income statement into the cash flow statement; working capital changes must reconcile with balance sheet movements; depreciation add-backs must land in the operating activities section. Done manually, this period-by-period verification consumed two or more days of analyst time per company — before a single formula was written.
Energent.ai became the pre-modeling inspection engine
- Loaded the raw company-facts JSON in its native format — no preprocessing required.
- Surveyed all available XBRL tags programmatically, narrowing dozens of overlapping entries to the usable subset for each statement line.
- Pulled the most recent annual values for core income statement, balance sheet, and cash flow lines, flagging gaps and period-coverage issues.
- Mapped all three inter-statement connections — net income flow-through, working capital reconciliation, depreciation add-back placement — against the actual filing data.
- Delivered a structured task plan with two concrete execution options for the modeling walkthrough.
No manual tag census. No period-by-period hand-tracing. No silent formula errors from a mismatched template.
How David Park runs it day-to-day
- Upload the company-facts JSON directly to an Energent.ai session.
- The agent inspects the file schema and maps present statement categories.
- The agent runs the XBRL tag survey and returns a curated list of populated annual-period tags.
- The agent extracts core annual values and surfaces company-specific reporting quirks.
- The agent maps all inter-statement connections and presents two execution options for the modeling walkthrough.
Filing-specific analysis replaced template assumptions
- Actual tag inventory, not assumed. The agent analyzed what this company's EDGAR file contained — not what a standard template assumes — so output reflected real tag usage and real period coverage.
- Quirks documented before formula work. Company-specific inconsistencies in tag usage and one-time reclassifications were surfaced and logged before any formula was written, not discovered as blocking mid-model errors.
- Connection-first workflow. Every inter-statement link was validated before the formula stage, shifting the analyst's work from reactive error-fixing to proactive structure verification.
- Native file, no intermediaries. The agent worked directly against the uploaded JSON — no vendor normalization, no preprocessing pipeline, no subscription locked to a single normalized data source.
Pre-modeling inspection collapsed from days to a single session
- The usable XBRL tag subset was identified from a file containing dozens of overlapping entries — the manual tag-survey step eliminated entirely.
- All three inter-statement connections were mapped and validated in-session against the actual filing data.
- Company-specific reporting quirks were documented in the structured task plan, not discovered as blocking formula errors mid-model.
- The analyst entered the formula stage with a verified connection map already in hand, with two data-grounded execution paths ready to select.
"I went into the model with a clean connection map instead of spending the first two hours just figuring out what I was working with." — David Park, Senior Analyst at Meridian Capital Partners