PDFImage.docx.eml.msg
Read directly
PDFs, images, .docx, .eml and Outlook .msg files are read as they are.
Every exhibit. Every page. Found.
Our AI reads all the evidence in a file, finds every detail with its volume and page, and assembles your hearing documents. All on your own servers.
Sample file
question
When was the formal notice sent?
Ready8,622 pages
Try it here.Click a card, type a question or open the commands. Fictional data: nothing leaves your browser.
The softwareProjecteur, installed on your workstation. Offline: nothing leaves this machine.
Chapter IConfidentiality
I.1Law 25 · 100% local processing
Law 25, respected by design.
No document, no question and no transcript ever leaves your firm.
Learn more
Your biggest risk with AI is sending your clients' personal information to someone else's servers. Here, that cannot happen: the tool runs entirely on your firm's own infrastructure, with no outside cloud service. Professional secrecy stays where it belongs.
The tool takes care of data transfers. Your firm's other obligations under Law 25 (privacy officer, policy, incident register) remain its own.
- 0transfers outside Québec
- 0cloud providers
- 0internet needed
100% local processing
The models run on your servers and work without an internet connection. Nothing travels online, not even your questions.
Law 25
No personal information is communicated to a third party or outside Québec: processing stays under your control.
Chapter IIBringing in a file
II.1Bringing in a file
A messy disclosure, fifteen formats. One file.
The tool reads every format in a disclosure and tells you, before you start, what it contains and how long the analysis will take.
Learn more
Disclosure rarely arrives clean. The tool reads PDFs, images, .docx, .eml and Outlook .msg files directly, and converts everything else on site (legacy .doc, .rtf, spreadsheets, iWork), with no outside dependency.
.doc→text
.rtf→text
.numbers→table
Converted on site
Legacy .doc, .rtf, spreadsheets and iWork files are converted with no outside dependency.
Other enginejμmvam
Qwen3-VL · localjanvier
Scanned pages
OCR runs locally on Qwen3-VL, chosen after a benchmark against PaddleOCR, which read “jμmvam” for “janvier” on degraded pages.
Vol. 12Transcript · 2023-03-08215 p.
P-40Report · 2021-11-0214 p.
REC-3Audio · 2023-03-087 h 42
Inventory
What is in the file, exhibit by exhibit, with its type, date and size.
Full analysis estimate3 h 40
Estimate
Before launching the full analysis, the tool tells you how long it would take. Then you decide.
« Réception provisoire »“Provisional acceptance”
French and English
Ask in one language, find the evidence in the other. Working translations included.
II.2Audio and video
Hours of recordings. Every segment timestamped.
Whisper transcribes locally and timestamps every segment.
Learn more
Audio and video go through Whisper, on your servers. The timestamp is the provenance: a lead that does not cite a real interval of the recording is rejected by the code.
Sample recordingREC_2023-03-08.wav
00:00:0001:55:3003:51:0005:46:3007:42:00
04:12:08
WHISPER · LOCAL
7 h 42Duration
3,118Timestamped segments
0Leads without a real interval
Segmentsstart → end
Rejected
Proposed lead: “the witness admits the delay”, citing 08:14:02 → 08:14:09.
The recording lasts 07:42:00. That interval does not exist: the code discards the lead.
II.3Sorting and renaming
Files in a heap. Sorted, renamed, reversible.
The tool sorts your files and renames them using the mode you choose: date, name, type or size. Every sort can be undone in a single step.
Learn more
Each file is renamed using the chosen mode, for example prefixed with its date or rank. It is reversible: defaire_classement restores the original names.
Mode:
BeforeDisclosure/
Sorted and renamedmode: date
Fictional example · 8 files sorted and renamed
Chapter IIIFinding and summarizing
III.1Finding a precise detail
The precise detail. With its volume and page.
Every result leads to the exact page of the original PDF.
Learn more
The link to the page appears on screen only: the deliverables handed to the firm never contain one.
question
Looks for a piece of information in the exhibits and returns the passage with its volume and page.
mentions
Which exhibits talk about this, and how many times. The question that needs no drafting.
partout
The same thing, across every file in the database at once.
chronologie
The dated moments exactly as stated in the evidence, with no model at all.
Accents123 passages
found for “harcelement” in a real file, even typed without its accent.
Pluralsbanderole(s)
Singular and plural are found together.
Speed7 ms
to count mentions across the whole file.
Demo · 12 fictional exhibits · Accents and plurals folded · No real data, nothing sent.
III.2Document summaries
Pages of testimony. The gist in three points.
A passage, an exhibit or an entire testimony, summarized in a few points.
Learn more
Every point in the summary refers back to its original lines: you check at a glance. Choose key points or a paragraph.
Vol. 12 · cross-examination, witness Bp. 97
01Q. Ma'am, you received the complaint on March 3?
02A. Yes, by email, late in the day.
03Q. What did you do next?
04A. I forwarded it to human resources the next day.
05Q. When did the investigation begin?
06A. A few weeks later, I believe.
07Q. Did you read the investigation report, exhibit P-40?
08A. No, I never read it.
09Q. Did the complaints register exist in March?
10A. I don't know.
11Q. Who was responsible for it?
12A. Human resources, I believe.
III.3Indexes from the front matter
The court reporter's indexes, read exactly as printed.
Table of contents, exhibits, objections and undertakings, read exactly as printed.
Learn more
At the start of every volume, the court reporter compiles these lists by hand. The tool reads them with no model: nothing can be invented.
Volume 12 · front matterp. i
Table of contents
Examination-in-chief of witness B3
Cross-examination89
Re-examination202
List of exhibits
P-40 Internal investigation report47
D-7 Prevention policy131
Objections
O-31 Relevance97
O-32 Hearsay160
Undertakings
U-12 Produce the complaints register118
Where each examination begins
Witness B · in chiefVol. 12, p. 3
Witness B · crossVol. 12, p. 89
Witness B · re-examinationVol. 12, p. 202
What each exhibit is, and where it was filed
P-40 · Internal investigation reportVol. 12, p. 47
D-7 · Prevention policyVol. 12, p. 131
Objections and undertakings
O-31 · RelevanceVol. 12, p. 97
O-32 · HearsayVol. 12, p. 160
U-12 · Complaints registerVol. 12, p. 118
On a real file
45volumes
8,622pages
50examinations
349numbered exhibits
84objections
98undertakings
0.1 sto extract it all
Chapter IVPreparing for the hearing
IV.1The four assemblies
What you used to build by hand the night before the hearing.
Four hearing documents, produced with no model at all.
Learn more
These are the documents a lawyer assembles by hand before a hearing. Every line comes from the indexes and pages of the file.
Assembly
Brings together the witness's volumes, phases, counted non-answers, and the exhibits and objections of their testimony.
| Section | Witness B | Where |
|---|---|---|
| Volumes | 3 transcript volumes | Vol. 12 to 14 |
| Phases | In chief, cross-examination, re-examination | Vol. 12, p. 3 · p. 89 · p. 202 |
| Non-answers | 6 in chief, 19 in cross | Page-by-page detail |
| Exhibits | P-40 D-7 P-52 | Vol. 12, p. 47 · p. 131 · Vol. 13, p. 22 |
| Objections | 4 objections | Vol. 12, p. 97 · p. 160 · Vol. 13, p. 58 · p. 141 |
| Where | P-40 · Internal investigation report | Context |
|---|---|---|
| Vol. 12, p. 47 | Exhibit filed, 14 pages | Examination-in-chief, witness B |
| Vol. 12, p. 52 | Questions on the report's conclusions | Examination-in-chief, witness B |
| Vol. 13, p. 8 | Back to the date of the investigation | Cross-examination, witness B |
| Vol. 21, p. 310 | Witness says they never read the report | Examination-in-chief, witness D |
| Undertaking | Subject | Given at | Status |
|---|---|---|---|
| U-9 | Send the March emails | Vol. 11, p. 204 | Followed up · Vol. 15, p. 4 |
| U-12 | Produce the complaints register | Vol. 12, p. 118 | Never followed up |
| U-13 | Provide the org chart | Vol. 12, p. 176 | Followed up · Vol. 16, p. 2 |
| U-17 | Check the timesheets | Vol. 14, p. 63 | Never followed up |
| Exhibit | Discussed at | Context | Declared in a list |
|---|---|---|---|
| P-58 | Vol. 19, p. 233 | Email shown to witness C | None |
| D-14 | Vol. 27, p. 61 | Policy excerpt read aloud | None |
| P-61 | Vol. 40, p. 12 | Photo discussed by witness F | None |
Fictional example · no model in the assemblyDeliverables without clickable links
IV.2Witnesses
A change in demeanour, measured rather than guessed.
Each witness's “I don't know” answers and answer length, measured in chief and in cross.
Learn more
temoins measures the change in demeanour between the two: the number of non-answers and the length of the answers. For non-answers, the threshold is a statistical test, the two-proportion z-test, not an arbitrary gap.
5 witnesses
flagged by the statistical test on a 45-volume file.
14 pts
gap for the first of them, between examination-in-chief and cross.
17 witnesses
identified from the printed phase headers.
Non-answers and answer length, in chief and in crossFictional example · threshold |z| ≥ 1.96
How to read it: a witness who answers “I don't know” far more often in cross-examination than in chief, and whose answers get shorter, has changed demeanour. The statistical test tells you whether the increase is too large to be chance.
| Witness | “I don't know” in chief | “I don't know” in cross | Increase | Answer length chief → cross | Verdict |
|---|
Identification
The witness is identified from the header printed at the top of every transcript page, the one source that does not lie.
credibilite
Surfaces what lets you challenge a witness: their counted non-answers and the passages that lend themselves to it, deliberately favouring exclusion over volume.
IV.3Employment law
A recourse deadline, calculated, never guessed.
The 45 days under section 123 ALS, calculated to the day.
Learn more
It is pure, repeatable arithmetic. An unknown date is treated as unknown instead of being guessed. The concurrency rules of section 123.4 and the thresholds come on top.
Deadline
October 1, 2026
August 17, 2026 + 45 days
Sample calculation, for illustration only.
Chapter VReliability
V.1Zero hallucination
Zero hallucination. Only your file.
The tool answers only from your documents. No source, no answer.
Learn more
General-purpose online AI answers from memory and can invent a quote, a date or an exhibit. Here, everything runs locally, without internet. The indexes are read exactly as printed, the timeline and assemblies are produced with no model, and an audio lead that does not cite a real interval is rejected by the code.
question
Did Ms. Gagnon admit the delay in writing?
General online AI
Yes. In an email dated April 5, 2021, Ms. Gagnon acknowledges that “the work is three weeks behind schedule because of her team”.
Exhibit not foundNo April 5 email in the file. Invented quote.
Lavoie&Maher
No exhibit in the file contains a written admission of the delay by Ms. Gagnon. The closest passage points the other way: she blames the delay “exclusively on the rain”.
D-2Vol. 2 · p. 21 · verified in the file
Where answers come fromThe memory of a model trained on the internetOnly the exhibits in your file
Answer without a sourcePossibleRefused
Evidence missingRisk of an invented answer“No exhibit establishes this”
Needs internetYes, at all timesNo, works offline
Where your data goesA third party's servers, often outside QuébecStays on your servers
Fictional example, for illustration only. Answers from online AI tools vary.
From file to hearing
From a messy disclosure to the hearing binder.
Bring in
Drop the disclosure in as it arrives. The inventory tells you what it contains, the estimate tells you how long the analysis will take.
Find
question, mentions, partout and chronologie, with the volume and page of every passage.
Assemble
Cross-examination binders, exhibit files, undertakings tracking and undeclared exhibits.
Private demo
Entrust us with a file. We will show you its evidence.
A 30-minute meeting with the Lavoie&Maher team, on a closed or anonymized file of your choice.