The Question Nobody Wants to Answer
It is 3:47 on a Friday afternoon. The owner's project executive emails: "Just confirming — what did we decide about the controls package on the OAC last week? Trying to close out the budget memo before the board meeting Monday."
If you are running a serious commercial project, you know exactly what happens next. The PM scrolls through 280 unread emails. They open the meeting minutes in Procore and find half a paragraph that says "Controls scope discussed; to be revisited." They text the foreman. The foreman remembers the conversation but is on a backhoe in another county. The mechanical engineer is on vacation. By 5:30 you have given the owner a hedged answer, requested a clarification call for Monday morning, and reshuffled your weekend.
"What did we decide?" is the most expensive question in construction. It is asked dozens of times a week on every active job, and the cost of answering it badly compounds across change orders, schedule slips, and erosion of owner trust. Yet none of the modern construction software stack actually answers it. Procore stores the meeting minutes. Sage stores the billing. Bluebeam holds the markups. Microsoft 365 holds the email. None of them connect.
That gap is what Project Pulse is built to close.
What “Cited” Means
Project Pulse is a retrieval-augmented generation layer that sits over every document the project produces. Specifications. Sheet drawings. Submittals. RFIs and their resolutions. Meeting transcripts with audio anchors. Email threads imported through the connector. Daily logs. Pay applications. Change orders. The data the other 27 engines emit as they run.
When a PM asks a question — "what did we decide about the controls package?" — Pulse retrieves the relevant slices of project memory, generates a plain-language answer, and pins the supporting citations next to it. Every citation is a deep-link: click it, jump to the source. The meeting transcript opens to the exact timestamp. The RFI opens to the exact resolution clause. The pay app opens to the line item being referenced.
Three numbers matter. The average Pulse answer takes 3.4 seconds. The median answer references 8 sources. And 100% of factual claims are cited — there is no narrative paragraph that says "Marcus committed to delivery on Thursday" unless Pulse can show you the audio clip of Marcus saying it.
Why “Cited” Is Not Optional
An uncited AI answer is a guess. On a $50M project, a guess gets repeated in a memo, the memo turns into a decision, and the decision becomes a $148k change order three months later. Pulse refuses to make claims it cannot back up. If the project memory does not contain an answer, Pulse tells you so — and tells you what document is missing.
A Real Example
Back to the Friday afternoon email. The PM opens Pulse, pastes the owner's question, and asks: What did we decide about the controls package on the most recent OAC?
Pulse answers in 3.1 seconds. It says: "At the OAC on April 23, the controls scope was reassigned to a different subcontractor effective immediately. Procurement tracker reflects ROJ delivery of June 14; submittal due May 27." Below the answer, four citations:
- Meeting · Apr 23 · 14:32 — audio-anchored. The PM clicks. The transcript opens at the timestamp. The foreman's voice plays: "We are moving the controls package to a different sub. Effective today." Twenty-two seconds of context, captured.
- Procurement tracker · ROJ field — last edit Apr 24. Click to view the tracker row. The PO date, the fabricator, the ROJ commitment, the submittal due date, all in one view.
- Submittals · Controls package — status: pending. Click to view what is being reviewed and what is outstanding.
- RFI 028 · Resolved Apr 22 — references the spec section that triggered the reassignment.
The PM copies the Pulse summary into a one-paragraph reply to the owner, adds the audio link, and sends. Total time from question to answer: under two minutes. The Friday afternoon is recovered.
Why Audio Changes Everything
Almost every consequential decision on a construction project happens in a meeting, and almost no meeting produces minutes that survive contact with a dispute. The PM took notes on a legal pad. The architect took notes in Procore. The sub remembered something different. Two weeks later, three people are sure of three different versions of the same conversation, and you are in the whose-notes-are-right argument with a $48k schedule cost on the line.
Pulse anchors to audio because audio is the only artifact that cannot be remembered selectively. When the Meetings · Voice engine records a session, it transcribes it, identifies decisions, auto-creates tasks from commitments, and pins each one to its timestamp. Pulse indexes all of it. So when the question is "did Marcus actually commit to Thursday?" — the answer is not a paragraph from a PM's recollection. It is the audio clip, playing.
That clip is in the project. It will be in the project ten years from now. The whose-notes-are-right argument does not happen.
The Compounding Effect
Pulse gets better at answering questions about your project the longer you run the project — because every engine output and every uploaded document expands the memory it draws from. A bid that gets leveled in week three becomes a citation source in week thirty. A submittal that gets reviewed in week eight becomes the reference the closeout team queries in week ninety.
This is the point of having 27 engines and 22 trackers running on one platform. Each engine produces structured output. Each tracker holds the live state. Pulse indexes both. So when a PM asks "where is the RTU?", Pulse does not search — it already knows. The Procurement tracker has the line in stage 5 (Shipped). The Submittal Reviewer has the approved submittal. The Schedule Risk engine has propagated the delivery date into the gantt. The bid that was leveled six months ago has the line item that originated this whole chain.
The output is operational. The PM gets one paragraph and four citations. The system does the joining underneath.
Try Pulse on Your Own Project
Drop a bid package, a meeting transcript, and a few RFIs into Trueleveler. Pulse will index the lot. By the end of your trial, ask it a question that is currently buried in someone's email. See the answer come back cited.
The Strategic Layer Underneath
There is a longer-term story under Project Pulse, and it matters more than the question-and-answer interface most users will see day to day. As projects run through the platform, Pulse is also building an anonymized, normalized procurement dataset — bids, scopes, change order patterns, pay application reconciliations, vendor performance — broken out by region, project type, and trade. That dataset compounds in value with every additional project on the platform.
Over time, the value is not "Pulse answers questions about your project." The value is "Pulse can tell you what a typical commercial fit-out actually costs per SF in the Pacific Northwest, what the median OH&P on a structural CO looks like in your region, and which scope items get under-bid 60% of the time." That is the pricing intelligence layer the entire construction industry has been missing.
This is why the Founding 25 cohort matters strategically. Founding members are not just locking in lifetime pricing — they are the first users to see the benchmark data as it builds, and they shape what gets indexed and surfaced as the dataset grows.
What Pulse Replaces
The "ask 5 people, get 3 answers" workflow. The PM weekend spent in email archaeology. The whose-notes-are-right argument. The dispute that turns into a $48k change order because nobody could prove what was said. The Friday afternoon reshuffle when the owner's question lands.
The Bottom Line
Construction has spent two decades digitizing documents. The documents are now searchable, version-controlled, and stored in the cloud. None of that helps when the owner asks "what did we decide?" — because the answer lives across five tools, three people's memories, and one meeting where nobody took good notes.
Project Pulse is the layer that closes that gap. It reads the documents the rest of the platform produces and answers the questions the platform was never able to. The answer comes back in three seconds, with citations, with audio anchors, and with the next action already taken if action was needed.
Drop the bid. The rest writes itself.
Try it on a real project
Trueleveler's Founding 25 cohort is still open. Twenty-five testers lock in $99/mo Pro pricing for life, get first access to the benchmark dataset as it builds, and have a direct line to the team to shape what gets indexed. Apply at trueleveler.com — 14-day free trial, no credit card.