How Melo's AI Works: Accuracy, Sources and What It Won't Answer
The first question every board asks is whether an AI will invent rules. Here is exactly where Melo's answers come from, how you verify them, and the questions it deliberately refuses.
Your documents only · Citation on every claim · Refuses when unsure
The Pipeline
From Documents to Intelligence
Your community's knowledge goes through a four-stage transformation before the AI can answer a single question.
Document Ingestion
CC&Rs, bylaws, meeting minutes, emails, handbooks — every document is parsed and prepared for processing. We handle PDFs, text, and structured data automatically.
PDF extraction → raw text → metadata taggingSemantic Chunking
Documents are intelligently split into meaningful passages — not arbitrary 500-word blocks. Each chunk preserves context so the AI understands complete thoughts, not fragments.
Context-aware splitting → overlap windows → boundary detectionVector Embeddings
Each chunk is transformed into a high-dimensional vector — a numerical fingerprint that captures meaning, not just keywords. 'Guest parking rules' and 'visitor vehicle policy' become neighbors in vector space.
text-embedding-3-small → 1536-dim vectors → pgvector indexingSemantic Retrieval
When a resident asks a question, we convert it to a vector and find the most semantically similar passages — even if they use completely different words. This is the magic of RAG.
Query embedding → cosine similarity → top-k retrieval → rerankingIn Real Time
What Happens When a Resident Asks a Question
Every question triggers a precise retrieval pipeline — here's the journey from question to cited answer in under 3 seconds.
Resident Asks
Natural language question
Embed Query
Convert to 1536-dim vector
Semantic Search
Find similar passages
LLM Generates
Compose cited answer
Cited Answer
Verified, sourced response
"Can I rent out my unit on Airbnb?"
Short-term rentals of less than 30 days are prohibited under your community's CC&Rs.[1] Owners may lease their units for periods of 6 months or longer with prior board approval.[2]
Accuracy
Six Safeguards Between a Question and an Answer
"Is AI accurate enough for HOA rules?" is the right question to ask. Accuracy here isn't a claim about the model — it's a set of design choices that make every answer checkable.
Retrieval before generation
The model never answers from memory. Relevant passages are retrieved from your documents first and passed in as the only permitted source material for that answer.
Citations on every claim
Numbered superscript citations tie each statement back to the specific document and passage, so any board member can verify an answer in one click rather than trusting it.
Financial questions get stricter sourcing
Budget and assessment questions are restricted to official adopted budget documents — never email threads or minutes — so the numbers residents see stay consistent with the board's approved figures.
Refusal is a feature, not a failure
The system is tuned to prefer "I don't know, let me route this" over a confident guess. In governance, a wrong answer costs far more than a missing one.
Amendments supersede originals
Upload an amendment and it takes effect immediately. Retiring an outdated document removes it from retrieval, so superseded rules stop being quoted.
Every answer is reviewable
Boards and managers can see what residents asked, what Melo answered, which sources it used, and where it declined — the full record, not a black box.
No AI system should be described as incapable of error. The honest standard is verifiability: because every answer shows the document and passage behind it, a board member can confirm or correct anything in seconds — which is more than can be said for the answers residents currently get from neighbors and old email threads.
Deliberate Limits
What Melo Won't Answer
The refusals matter more than the features. These are scoped out on purpose — they involve judgment, legal exposure, or fiduciary duty that belongs to your board.
Whether a rule was violated
Melo will tell a resident what the parking rule says. It will not decide whether their neighbor broke it. Enforcement involves judgment and due process, and that stays with the board and manager.
Legal interpretation of your documents
Questions about statute, fair housing, enforceability, or what a clause "really means" are routed to a person. Melo quotes the document; it does not opine on it.
Anything about an individual's account
Balances, delinquencies, liens, collections, and payment disputes are outside the assistant's scope entirely. Those conversations belong with management.
Board decisions that haven't been made
If the board hasn't voted on something, Melo won't speculate about the outcome or imply a position. Silence in the documents is reported as silence.
Questions your documents don't cover
This is the important one. When retrieval finds nothing relevant, Melo says it doesn't know, offers to escalate, and logs the gap for the board — instead of producing a plausible-sounding guess.
General internet knowledge about HOAs
Melo is scoped to your community's uploaded documents. It will not answer using another association's rules or generic web content that has nothing to do with your CC&Rs.
What happens instead: the question is handed to the right person — operations questions to management, governance questions to the board — and logged so you can see the pattern over time. Nothing quietly disappears.
An AI That Gets Smarter Every Day
Most AI tools are static. Ours evolves. Every unanswered question is an opportunity, and every official answer makes the model more accurate.
FAQ Training
3 unanswered · 12 trained
"What are the pool hours during the summer?"
Asked 8 times this week
"How do I submit a maintenance request?"
Trained · AI answers correctly
"What's the guest parking policy?"
Trained · AI answers correctly
The Feedback Loop That Eliminates Repeat Questions
When the AI encounters a question it can't answer from existing documents, it doesn't guess — it flags it. Your team provides the official answer once, and the model learns instantly. Over time, the percentage of unanswered questions shrinks to near zero.
Answer accuracy over time (representative)
Everyone Contributes
Three Stakeholders, One Smarter AI
The best part? Nobody needs to be technical. Each stakeholder naturally makes the AI smarter through their regular interactions.
Homeowners
Ask questions → Surface gaps
Every unanswered question is a signal. Resident interactions reveal exactly what information is missing from the knowledge base.
Board Members
Approve policies → Train accuracy
Board decisions and approved policies become training data. Official answers ensure the AI reflects current governance, not outdated rules.
Management
Provide answers → Monitor quality
Your team provides official answers to frequently asked questions. Each response instantly trains the model, reducing future support volume.
Trust & Accuracy
Why This Isn't Just Another Chatbot
Generic AI makes things up. Our architecture makes that physically impossible.
Strict Sourcing Policy
Every answer must be grounded in your community's documents. If the AI can't find a source, it says 'I don't know' instead of guessing.
Transparent Citations
Numbered superscript citations link every claim back to the specific document and passage. Residents can verify any answer instantly.
Grounded, Not Guessed
Unlike generic chatbots, answers are generated only from passages retrieved from your uploaded documents — and if nothing relevant is found, the AI declines instead of inventing a policy.
Always Current
Upload a new document or update existing ones and the AI's knowledge updates immediately. No retraining delays, no stale information.
Accuracy and Sourcing Questions
The questions risk-averse boards ask before they say yes.
Is AI accurate enough for HOA rules?
It depends entirely on architecture. A general chatbot answering from training data is not appropriate for governance, because it will produce fluent answers about rules it has never seen. A retrieval-based system that quotes your own uploaded documents and cites the passage is accurate in the way that matters: every claim is checkable against the source in seconds.
Does Melo hallucinate HOA rules?
Melo generates answers only from passages retrieved from your community's documents, and each claim carries a citation you can open. If retrieval returns nothing relevant, it declines rather than inventing a rule. The practical safeguard is verification: because sources are always shown, any inaccuracy is visible immediately instead of hidden in confident prose.
Where do the answers come from?
Only from documents your community uploads — CC&Rs, bylaws, recorded amendments, rules and regulations, architectural guidelines, adopted budgets, amenity policies, minutes and newsletters. Nothing is answered from the open internet or from another association's documents.
What happens when the documents don't have the answer?
Melo says so plainly, offers to route the question to the right person, and records it as a documentation gap. Boards review that list monthly and use it to decide which documents or policies to update.
How does it handle conflicting or outdated documents?
Recency and document type are used to prefer current governing versions, and citations show exactly which document an answer came from so conflicts are visible rather than silently resolved. Retiring superseded documents removes them from retrieval entirely, which is why document cleanup is the first step of any rollout.
Can the board see what residents are being told?
Yes. Boards and managers can review questions, answers, the sources used, and unanswered questions. Nothing about the assistant's behavior is hidden from the people accountable for the community.
Is our community's data used to train shared AI models?
No. Your documents are used to answer your community's questions. They are not pooled into a shared model that other communities benefit from.
Can Melo issue violations or make decisions for the board?
No, by design. Melo explains what documents say. Determining a violation, approving an architectural request, or setting policy remains a board function.
See It In Action
Book a demo and we'll show you exactly how Melo's AI processes your community's documents and delivers accurate, cited answers.
Personalized walkthrough with your own documents