10 June, 2026 - Last week in TI
- Rahul Subramaniam
- Releases
- June 10, 2026
Mosaic: Turn an Idea into a Finished Video — Now Live
Mosaic is now live. Describe a video idea in plain language and Mosaic builds a complete, narrated video for you — writing the script, designing the characters and scenes, and rendering the final cut. Work through it stage by stage, or hit Auto and let it run end to end.
- Live and ready to try — Sign in and turn a single sentence into a finished video at mosaic.ti.trilogy.com.
- Idea to video, guided — Your concept moves through Idea, Research, Concept, Script, Assets, Scenes, and the final video — each stage builds on the last and stays editable.
- Auto mode with email when ready — Hit Auto and Mosaic runs the full pipeline on its own. Close the tab and walk away; it keeps working on the server and emails you when your video is ready to review.
- Better storytelling and consistency — Improved narration and pacing, more consistent characters and scenes across clips, and background music for a more polished result.
- Plan before you generate — Each clip gets a storyboard preview of the shot so the scene is planned before any video is generated, and an AI Director in the editor lets you refine or regenerate clips in plain language.
Athena Applets: Strong Approval Week & Grade 6 Expansion
43 lessons approved this week with a 32% approval rate. 1 new Grade 6 lesson generated. Total review time was 999.3 minutes (7.5 minutes per iteration). The autonomous agent resolved 349 out of 359 identified issues (97% success rate) across ~55 unique lessons.
- Production output – 43 lessons approved (Grade 3: 35, Grade 3 Supporting: 8). 1 new lesson generated and entered review queue (CCSS.MATH.CONTENT.6.EE.C.9+2, Grade 6).
- Approval rate – 32% approval rate (43 approvals from 133 total iterations across 77 unique lessons).
- Autonomous agent performance – 97% success rate (349 issues resolved, 10 unresolved) across ~55 unique lessons. 28 additional lessons were skipped due to Drive-lock contention.
- Efficient iteration cycle – 133 total iterations completed in the past 7 days (90 feedback rounds, 43 approvals) with an average of 4.5 comments per lesson (348 total comments across 77 unique lessons).
- Review queue – 79 lessons ready for validation (Grade 3: 2, Grade 3 Supporting: 15, Grade 6: 57, Grade 6 TEKS: 2, Grade 7: 3).
- Review time insights – Average review time of 7.5 minutes per iteration (total 999.3 minutes).
- Complete lesson catalog – See the full list of all uploaded lessons across grades in this lesson catalog with direct links to each lesson.
EduPaid v2.30.1: Payment Receipts, Reliability & Platform Improvements
EduPaid v2.30.1 delivers direct Stripe receipt links in confirmation emails, end-to-end payment reliability fixes, statement delivery corrections, and a modernised platform with proactive monitoring — so families spend less time chasing records and support.
- Receipt links in confirmation emails — Families now receive a direct link to their Stripe receipt the moment a payment completes. The link opens the official Stripe receipt with full payment details and can be forwarded or shared as proof of payment — no account hunting, no support tickets.
- Payment processing reliability — Resolved an issue where some payment events did not process correctly after completion. Payments now flow through reliably from start to finish. No action needed — any affected payments have already been corrected automatically.
- Monthly statement delivery — Fixed an issue where monthly statement emails were not reaching the correct recipient in some cases. All statement emails now deliver as expected.
- Faster, more reliable platform — Several core services have been migrated from legacy infrastructure to a modernised, unified backend — reducing complexity and improving response times. System-wide monitoring and automated alerts are now live, so issues are detected and resolved before they surface to customers.
EduLLM SAT (Reading and Writing)
EduLLM SAT-RW achieved significant progress with base training reaching ~98.6% performance while addressing persistent standard failures. Parallel ranker development and PSAT-10 optimization efforts are underway, with PSAT-10 reaching a peak of 92% performance.
- Peak Base Training Performance – SAT-RW reached approximately 98.6% with the base training layer while focusing efforts on standards that consistently fail to improve overall reliability.
- Ranker Development in Progress – Working on ranker implementation in parallel, though initial ranker integration showed no improvement over the base model.
- PSAT-10 Optimization – Dedicated time to PSAT-10 improvements, achieving a peak score of 92% and identifying areas for further enhancement.
Marauders Map: Safety Dashboard & After-Hours Monitoring
Two major Safety & Security capabilities this week: a real-time Safety Dashboard giving admins a single-screen command centre for campus status, and an end-to-end After-Hours Monitoring pipeline that detects, notifies, and verifies people on campus outside school hours.
- Safety Dashboard — A single screen showing live campus status (Clear through Emergency), headcount and camera stats, room-by-room presence with student and guide counts, an alert feed flagging items that need attention, and a one-click Emergency activation panel — all updating in real time.
- After-Hours Monitoring — Automatically detects anyone on campus outside school hours, sends email alerts to admins, and provides a dedicated tab where each detection can be reviewed and marked as Authorized, Unauthorized, or False Alarm — with a full audit trail of every decision.
Alpha Coach Analysis Platform: Smart Coach Scheduling & Calendar Sync
Scheduling coaching sessions used to mean juggling spreadsheets and manual calendar invites. This week we built a scheduling engine for Alpha Coach that matches each student to the right coach automatically and handles all the booking and calendar work behind the scenes.
- The right coach, every time — Coaches are tagged by grade, subject, and the type of call. When a student needs a session, they’re automatically matched with a coach who fits — no manual sorting, no guesswork.
- Calendars that stay in sync — Bookings flow straight into coaches’ Google and Microsoft calendars, with a video link and one-tap add-to-calendar in every confirmation. Coaches see their whole day in one place, and double-bookings disappear.
- Set it once, runs itself — Admins build a session template once and roll it out to every coach with the key details locked in. Coaches set their own hours, and school holidays and days off import in a single click.
- Book from anywhere — Students can book a session right inside Timeback — no extra login, no redirect, no friction.
BrainTrust: Faster, Smarter Agent Execution with LangSmith Sandboxes
BrainTrust’s backend has been fully migrated to LangSmith sandboxes, delivering dramatically faster performance on data-heavy tasks.
- Up to 3x faster execution — Filesystem and code execution operations now run significantly faster, making large-scale data tasks feel near-instant.
- Finance agents can now create and edit more document types — Support now covers Simple Journals, Advanced IC Journals, Credit Memos, Customer Payments, Deposits, and Direct Invoices.
- 65.6% cost reduction via prompt caching — Prompt caching cuts the cost per 1M tokens from $3.11 to $1.07 — making BrainTrust roughly 2.9× cheaper to run.
- Coming soon: Subscriptions, billing accounts, and invoices — Finance agents will soon support creating and managing Subscriptions, billing accounts, and invoices and credit memos via subscriptions.
EduLLM SAT Math: Strong Scores Across All Three Grades
SAT Math posts strong results across all three Digital SAT grades. The only remaining gap to 99% is difficulty alignment — new training examples are being generated to fix this.
- Digital PSAT 8/9 – 97.78% pass rate, 99.23% aggregate, 80.65% variety (270 evaluated items).
- Digital PSAT/NMSQT and PSAT 10 – 95.10% pass rate, 98.74% aggregate, 86.57% variety (204 evaluated items).
- Digital SAT – 94.74% pass rate, 98.96% aggregate, 83.68% variety (228 evaluated items).
- Remaining gap – Only failure reason is difficulty alignment. New training examples are being generated to resolve this and push all grades to 99%.