Overview

Elfworks is a digital workflow platform serving property valuers across Australia, created to move firms off hand-edited Word templates and onto a guided, auditable process that carries a job from creation through to a signed report. Rather than treating a valuation as a document to be written, the platform treats it as a sequence of steps, capturing on-site measurements, condition notes, site observations, property photographs and market commentary as structured records. Bacancy Technology built the workflow engine behind that sequence, an AI-driven document pipeline capable of reading title reports and council records even when they arrive as scans, and one-step generation of the finished Microsoft Word and PDF report, together with the multi-tenant business layer spanning organizations, projects, users, job lifecycles and Stripe billing. Valuers now spend their time reviewing figures instead of retyping them, reports leave the firm in a consistent structure, and the numbers that carry professional liability are entered once.

Technologies Used

Project Highlights

checkmark

Unified inspection data, market research and source documents into a single guided valuation workflow.

checkmark

Automated Microsoft Word and PDF report generation directly from captured job data.

checkmark

Delivered AI document extraction with OCR and vision fallback for scanned property records.

checkmark

Integrated multi-provider AI routing with automatic failover across four model providers.

The Challenges

1

Every valuation report was assembled by hand inside a Word template, and the same figures were keyed in more than once. A measurement taken on site went into inspection notes, then the report body, then the summary. Two valuers at one firm produced differently structured documents, and each re-entry risked transposing a number carrying real liability.

2

The source material a valuation rests on arrives as PDFs. Title reports, land reports, zoning documents, comparable sales and council records are more often scans than digital originals, so a pipeline written on the assumption of readable text fails quietly across a large share of the intake and hands the work straight back to the manual entry it was built to remove.

3

Because document extraction sits on the critical path of every job, depending on one AI vendor meant depending on one point of failure. Throttling, quality drops or an outage would have stopped valuers mid-job, and a single-provider commitment would have locked the platform into that vendor's pricing and rate limits with nowhere to move if either changed.

4

Property valuation is a regulated act of professional judgment, not a calculation. Had the platform allowed a model to produce the final figure, the firm would have faced a defensibility problem the first time a valuation was disputed, so the automation had to compress every task surrounding the decision while leaving the decision itself entirely untouched.

Solutions by Bacancy

1

Our RPA development services team retired template editing in favour of a structured workflow engine. Every step writes into a defined data model, and the report generator renders that model into a firm-standard Word and PDF document on request. Inspection figures and comparable sales travel from capture to finished report without anyone retyping them, and every report the firm issues now shares one structure and one audit trail.

2

We built a document pipeline that first works out whether an uploaded file holds real text or is only an image, then routes it accordingly. Files carrying text go straight to extraction, while scanned originals pass through OCR paired with vision-enabled AI. Both routes return standardized business fields rather than an undifferentiated block of text, which is what makes the downstream workflow automation possible.

3

Our Python developers built multi-provider AI routing spanning Anthropic Claude, OpenAI GPT, Google Gemini and xAI Grok, failing over the moment a provider becomes unavailable, with response caching so repeat document work is never billed twice. Reliability and unit economics both improved from one architectural call made ahead of the first outage rather than in response to it, which is where this kind of decision belongs.

4

We set the automation boundary at professional judgment and enforced it in code rather than in a policy document, because automation delivered into a regulated field has to speed up the work surrounding a decision without ever reaching the decision. AI handles research, document comprehension and workflow execution, while the research assistant gathers supporting information and final calculations stay with the qualified valuer.

Core Features

checkmark

Guided valuation job creation with inspection capture for measurements, condition notes, site observations and photos

checkmark

AI extraction from title, land, zoning and council documents, with text-versus-scan detection and OCR vision fallback

checkmark

One-step Microsoft Word and PDF report generation with summaries, approvals, digital signatures and standard formatting

checkmark

Multi-tenant organization, project and user administration with Stripe payments and central configuration control

No. of Resource

05

No. of Resource

Time Frame

March 2025 - Ongoing

Time Frame

Project Snapshot

Elfworks

Outcomes

1 guided workflow replaced hand-edited Word templates across the entire valuation process

3 rounds of manual re-entry per job collapsed into a single structured point of data capture

2 extraction paths, text-native and OCR with vision, brought scanned council records in scope

4 AI providers behind one extraction service, with failover through any single-vendor outage

1-step Word and PDF generation removed manual report formatting from every finished job

0 AI involvement in the final valuation figure, leaving every regulated decision with the qualified valuer

Technical Stack

BACKEND Python
FRONTEND ReactTypeScript
DATABASE PostgreSQL
AI PROVIDERS Anthropic Claude OpenAI GPTGoogle GeminixAI Grok
DOCUMENT EXTRACTION OCRVision AIText/image detection
DOCUMENT GENERATION Microsoft Word PDF
PAYMENTS Stripe
PROPERTY DATA SOURCES RP DataTitle and land registriesLocal government records
MULTI-TENANCY OrganizationProject and user scoping
AI COST OPTIMIZATION Response cachingDynamic provider routing

Experience With Bacancy

2500+ Projects Experienced Innovation with Bacancy!

Get access to an experienced team of developers and engineers from Bacancy, handpicked to ace your goals. Kickstart within 48 hours, no-risk trial.

Book a 30 min call

14+

Years of Business Experience

1458+

Happy Customers

12+

Countries with Happy Customers

1050+

Agile Enabled Employees