Forty hours of bodycam arrives in your inbox on a Thursday. Add the dashcam, the interview-room camera, three jail calls, a 911 recording, and a 60-page police report, and you are looking at a discovery dump that will eat your week if you let it. The instinct is to open the first video file and start watching. That is the slowest possible path. Bodycam files arrive with names like X13qrt9__x3455, no scene labels, and no indication of which officer was where. Watching linearly means re-watching the same traffic stop from three angles while the one 40-second clip that contradicts the report sits at hour 31. Here is the workflow I use to get through a dump this size in a day or two, with citations I can actually put in a motion.
Key points
- Sort the discovery dump by source before you watch anything: bodycam, dashcam, interview-room video, 911 audio, jail calls, police reports, and photos each get reviewed differently.
- Get diarized, word-level transcripts first so you can search 40 hours of footage in minutes instead of scrubbing timelines.
- Run every new file against the police report and earlier statements to surface candidate inconsistencies (report vs bodycam, statement vs jail call) as review signals you check by hand.
- Build a suppression-motion checklist (consent, search scope, stop duration, K-9, Miranda) directly from quoted, cited source moments.
- Keep AI output marked for human review and keep originals immutable; redactions are derived files with an audit trail, which matters for discovery turnover.
Group the evidence by source type first. Bodycam and dashcam from the same event, interview-room footage, 911 audio, jail calls, the police report PDF, and any photo exhibits each get a different review pass. This matters because they answer different questions. Bodycam shows what the scene actually looked like and what the officer actually said. The police report shows what the officer wrote later. Jail calls show what your client said when nobody was reading them their rights. 911 audio shows the complainant's tone and the timeline of the call-out. If you blur them together into one pile, you lose the ability to compare them against each other, and comparison is where the good impeachment material lives. Create a case, upload everything (video, audio, images, PDFs up to 2 GB each), and let each file land in its bucket. If the state sends attachments to a case-specific inbound email address, they ingest automatically and show up in the activity feed so nothing gets lost in a personal inbox.
You cannot review 40 hours of footage by watching it. You review it by searching it. Get diarized transcripts with word-level timestamps across all the video and audio first. Speaker diarization separates Officer A from Officer B from your client, and word-level timestamps mean every search hit points to the exact second in the exact file. When the transcript labels the speakers wrong, rename the speaker once and the fix propagates across every segment in every file. For the scanned PDFs and photo exhibits (the report with no text layer, the receipt photo, the screenshot of the text thread), run OCR so the text becomes searchable and citable while keeping bounding boxes for later redaction. Now the review pass changes shape. Instead of scrubbing, you search: "consent," "step out," "you're free to go," "stop resisting," "do you want a lawyer." Every hit is a moment you can open, verify, and cite.
The highest-value review work is comparing sources against each other: the police report narrative against the bodycam, the client's statement against the jail call, the complainant's 911 story against what they told the second officer on scene. You can do this by hand, tab by tab. It is slow and it misses things at 2 a.m. A faster approach is to have each new file compared against the earlier evidence in the case as it lands, with candidate inconsistencies filed as review signals before you ask. A report-vs-bodycam mismatch on the duration of the stop. A statement-vs-call-audio mismatch on who was present. The signals come in unreviewed and marked, and you verify each one against the source yourself. Treat these the way you would treat a co-counsel's tip: a lead to check, never a conclusion. Every signal should carry both sides' quotes so you can see the conflict in the source language before you decide whether it is real.
Once a review signal holds up, the next question is what to do with it. For a preliminary hearing or a suppression hearing, that usually means questions. Draft questions from the two quoted sides of the conflict, with every question tied to a quoted source. Keep them as questions. The moment you write "the officer is lying about the stop duration," you have made a credibility conclusion that is yours to argue in court, on your own record, not something to paste from a draft. Save the question list as attorney-review notes and revise it in your own voice before it goes anywhere near a witness. The same discipline applies to a motion to suppress. Sweep the record for consent, search scope, stop duration, K-9 deployment, and Miranda, and draft a motion outline with both sides quoted and cited. You still write the argument. The outline just guarantees every factual assertion in your motion points to a timestamped moment you have personally watched.
When bodycam, dashcam, and interview-room footage cover the same event, put them on one clock. Synced playback with an interleaved transcript lets you see what Officer B was doing while Officer A was asking for consent, which is often where the suppression argument lives. Auto-sync proposals computed from word-timing correlation (for example, 31 word matches with a 0.2-second spread) save you the manual frame-matching work. Check the match evidence, approve or reject, and never let a sync apply itself. From there, build the case timeline: manual events plus AI-suggested events, tasks, follow-ups, and named-party records for officers and witnesses with their aliases, so mentions across notes and signals line up.
Turnover often means redaction: bystander faces, minor victims, license plates, PII on screens, and the Social Security number read aloud on the jail call. Use frame-accurate video redaction (drag the rectangle, choose black or blur), word-level audio redaction (pick the exact spoken words and that span is muted), and image and PDF redaction. Every redaction produces a new derived file. The original stays immutable and every action is logged in the audit trail, because at some point opposing counsel or the court will ask what you changed and when. AI-suggested regions for faces, plates, and PII screens should file as suggestion cards marked for review. Nothing gets auto-applied. If a redaction tool silently blanks a face it guessed at, you have altered evidence without a record of deciding to.
Most of the tools in this space were built for someone else. Police-media redaction studios sell to agencies and FOIA offices. Prosecutor-side evidence platforms hand you a slow player and no analysis. Generic transcription services miss specialized legal terms. Research and memo AI has no media review at all. The free stack of local Whisper plus NotebookLM works on one clip and breaks at case scale: no diarization quality control, no citations, no redaction, no audit trail. Evidence Copilot is the alternative built for the defense side of the case. Every AI answer cites the exact source moment it came from, every AI output is marked for human review, and the product makes no legal conclusions. There is a read-only MCP connector so you can query cases from Claude Desktop, ChatGPT, Claude Code, or Cursor through a standard sign-in flow with no tokens to manage, and no defense-side competitor offers that. Pricing is published with no sales wall: Solo at $39 per user per month with 15 media-processing hours, Firm at $99 per user per month with 60 hours, and a PD Office tier at $750 per attorney per year with 100 hours. Overage is metered and hard-capped at three times the included hours, so a surprise 200-hour dump cannot overbill you. Every plan includes the full workspace; there are no feature tiers. New accounts get a 7-day full-access trial without a card, and every paid plan carries a 14-day money-back guarantee.
The dump is still 40 hours. Nothing makes the underlying evidence smaller. What changes is how much of it you personally have to scrub through: sorting by source, searching word-level transcripts, letting cross-source conflicts surface as leads you verify, and keeping every AI-assisted step marked for your review with the originals untouched. You end up at the preliminary hearing or the suppression hearing with a timeline, a question list, and a motion outline where every sentence points to a second of video or a page of the report. That is the part that wins. If you want to see the workflow on your own discovery, book a 20-minute walkthrough at evidencecopilot.com/book or start on the Solo plan at evidencecopilot.com/signup.