Customer profile
The team is a small architecture and facilities planning practice handling institutional food-service projects — cafeterias, dining halls, and commercial kitchens for public and corporate clients. Principals and project managers split time between AutoCAD design work and downstream handoff tasks: area tabulations, room schedules, and specification documents that need to land in Excel or Word for client review, cost estimating, and permit submissions.
The project in this case study was a 208-person dining facility with more than 20 distinct functional zones: hot and cold kitchen production areas, fish, vegetable, fruit, and baking prep stations, dry and refrigerated storage, refrigeration chambers, staff changing rooms, shower rooms, locker rooms, an office, a switchboard space, and a ventilation chamber. Room schedule accuracy matters at handoff — errors in area figures flow downstream into construction estimates, code compliance reviews, and interior fit-out tenders. A single transposed decimal in an area column can cascade through a cost plan before anyone catches it.
Problem
The immediate task was straightforward on paper: export the room schedule embedded in the bottom-right corner of the facility's DXF drawing into an Excel file for client delivery. In practice, the source file presented several compounding problems.
The DXF contained 25,123 entities — a mix of LINE, INSERT, ARC, LWPOLYLINE, HATCH, TEXT, MTEXT, and DIMENSION objects. The room schedule existed only as text geometry scattered across the drawing canvas, not as a structured database object or a dedicated schedule layer. There were no reliable column headers for the room-number column; the numbers appeared in the schedule body but without a matching header label in the extracted text region, leaving reconstruction to coordinate-based inference rather than simple labeling.
Manual approaches were slow and error-prone. Opening a 25,000-entity file in AutoCAD to copy-paste a 36-row table into Excel took meaningful time and introduced transcription risk: misread numbers, missed rows, and dropped decimal separators. The area values in the source drawing used inconsistent formatting — some cells read as "17,5m²" and others as "17,50m²" — meaning even a careful manual export would leave non-numeric text in a column that any SUM formula would silently skip.
Three rooms (101–103) sat at the top edge of the schedule region, where a spatial filter would typically clip them. A subtly incomplete handoff — a schedule that starts at room 104 instead of 101 — might go undetected until the client or cost estimator noticed the gap.
Why now
Facilities handoff timelines compress at the end of design phases. When a cafeteria project moves from schematic design into construction documents, the owner and general contractor both need room schedules, area tabulations, and finish specs quickly. Delays in producing a clean Excel handoff — even a day while someone manually transcribes a 36-row CAD table — create bottleneck pressure on estimating and permitting milestones that have fixed external deadlines.
For a small practice, there is also no dedicated BIM coordinator or CAD technician to absorb this extraction work. The architect or project manager who drew the plan is often the same person exporting the schedule. Time spent on transcription is time not spent on design or client communication. Automating the extraction, even for a single drawing, reduces per-project overhead in a way that compounds across a portfolio of food-service commissions with similar documentation requirements.
Why energent.ai
The team needed a tool that could accept a DXF file directly, reason spatially about the layout of text entities, and produce a clean Excel output — without requiring a custom script, a BIM plugin license, or a spreadsheet macro rebuilt for every project.
Alternatives fell short in predictable ways. AutoCAD's built-in data extraction wizard works cleanly when schedule data lives in structured block attributes; it does not handle tables drawn as plain text geometry. Spreadsheet-based OCR tools can process PDF exports from AutoCAD but introduce an additional export step and degrade accuracy when drawing geometry overlaps text. Hiring a CAD technician to transcribe the table manually was the default path, but it added cost and turnaround time for a task that should, in principle, be deterministic and verifiable.
Energent.ai accepted the raw DXF, loaded it without preprocessing, and used coordinate-based reasoning to isolate the bottom-right schedule region from the rest of the 25,000-entity drawing. It ran Python-based extraction logic directly against the DXF geometry, reconstructed the table row by row from spatial position, inferred the missing "Number" header from the structure of the data, and produced a formatted Excel workbook — all within a single session. An independent audit subagent then re-extracted the schedule from scratch and verified the output before final delivery.
Workflow
Step 1 — File ingestion. The architect uploaded the DXF file directly to energent.ai. The agent validated the file immediately: AutoCAD R2018 format (internal version AC1032), two layouts detected ("Model" and a watermark layout), 25,123 modelspace entities spanning LINE, TEXT, MTEXT, INSERT, HATCH, ARC, LWPOLYLINE, and DIMENSION geometry types.
Step 2 — Drawing context assessment. Before narrowing focus, the agent surveyed the full drawing. It identified the building type (208-person dining facility), catalogued more than 20 named functional zones across kitchen production, storage, and staff areas, and confirmed that the room schedule was embedded as text geometry in the bottom-right corner rather than in a structured block or table object.
Step 3 — Region isolation and text extraction. The agent extracted all text entities with their X/Y coordinates, then filtered to the bottom-right cluster that contained the schedule. It reconstructed the table row by row from spatial position — matching room numbers, room names, and area values by their relative coordinates — rather than relying on CAD layer names or object type.
Step 4 — Header recovery. The agent located column headers for "Room" and "Area, m²" in the drawing text. Because no explicit header text appeared for the room-number column, the agent applied "Number" as the Excel header based on the column's content and position. The subsequent audit confirmed this was appropriate given the schedule structure.
Step 5 — Completeness check and second pass. The initial extraction region filter clipped rooms 101–103 at the top of the schedule. The agent detected the gap, widened the vertical extraction bounds, and regenerated the workbook without user intervention. The final file contained all 36 rows, numbered 101 through 136.
Step 6 — Excel export. The agent produced the workbook with a sheet named "Room Table," a yellow-highlighted header row across all three columns, and 36 data rows in drawing order. Area values were preserved as-is from the source — "17,5m²" and "17,50m²" — so the architect could decide how to normalize them for downstream calculations rather than having the agent silently override source formatting.
Step 7 — Independent audit. A subagent independently re-extracted the schedule from the source DXF without relying on the primary agent's logic, then compared every row against the Excel file. Verdict: PASS. No missing rows, no extra rows, no duplicates, no ordering issues across all 36 entries.
Results
- 36 rooms extracted, covering the complete range from room 101 to room 136 with zero omissions in the final delivery.
- 37-row Excel workbook produced (1 yellow header row + 36 data rows) with correct column structure, sheet naming, and header formatting ready for client handoff.
- 3 rooms recovered (101–103) that the initial region filter missed — the agent self-corrected without user intervention rather than delivering a silently incomplete table.
- Independent audit passed with an exact match on room number, room name, and area text across all 36 rows — no discrepancies found between the Excel output and the DXF source.
- Manual transcription eliminated for a 36-row, 3-column table embedded in a 25,123-entity drawing, removing the risk of copy-paste errors propagating into area figures used for cost estimation.
The one remaining task for the team was area value normalization: the source drawing's mixed formatting means the area column cannot be summed directly in Excel until values are converted to numbers. Energent.ai flagged this explicitly and offered to add a normalized numeric column as a follow-on step — a clean handoff point where the architect retains control over decimal separator conventions and unit handling.
Proof
"The drawing had 25,000 entities and the schedule was just text scattered in one corner — there was no clean layer to export from. The agent found the table, caught the three missing rooms at the top, and handed me a verified Excel with a yellow header row. The audit result was what sealed it for me: a second pass that independently confirmed every room number and area against the DXF, not just the agent asserting it looked right."
— Architect and project lead on the dining facility handoff
The delivered workbook contains the full 36-room schedule on a sheet named "Room Table," with columns "Number," "Room," and "Area, m²," sorted in drawing order from room 101 to room 136. The independent audit confirmation — PASS, with no missing, extra, duplicate, or misordered rows — is part of the session record and available for project documentation.
Trust note
Area values in the Excel output are text strings, not numbers, because the source DXF used inconsistent formatting that the agent preserved rather than silently normalizing. Before using the "Area, m²" column in SUM formulas or area-total calculations, the team should convert values to numbers and confirm that decimal separator conventions match the project locale (comma vs. period). The "Number" column header was inferred from column structure rather than extracted from an explicit label in the drawing — appropriate for this file, but worth cross-checking against the project's room-schedule standard before submitting to a client or an authority having jurisdiction over the permit set.