We underestimate.
People reliably predict tasks will take less time than they do. Psychologists call it the planning fallacy (Buehler, Griffin & Ross, 1994).
Tempo learns your personal time multiplier per subject from the sessions you log.
AI for Student Life · a study planner for high school
A week you can actually finish.
Tell Tempo your week the way you'd text a friend. It learns how long work really takes you, spreads out your revision, keeps clear of practice and sleep, and tells you on day one when a week doesn't fit.
- Free, no account
- Works offline
- Private local AI option
Live The real planner, in your browser
3 assignments · 2 exams · robotics + football ·
3AI engines you can switch between: local, cloud, or none
22automated tests, covering the judges' scenarios and every bug we fixed
0accounts, emails or real names. Data stays in your browser
1 tapto tell Tempo you're tired. Today's load halves and it shows what moved
01 · The problem
Planners store your week.
None of them tell you if it fits.
We cram.
Spreading study over several days beats one long session the night before (Cepeda et al., 2006), but cramming is what usually happens.
Tempo splits exam prep across days: overview, active recall, then a practice test.
We find out too late.
An impossible week looks fine on Monday. You only discover it on Wednesday night, when there are no good options left.
Tempo spots it on day one and offers options it has already checked: more time, an extension, or a focused version.
02 · How it works
Type it. Check it.
Let the plan keep up with you.
-
"chem test next friday, english essay due monday about 2 hours, robotics tue and thu 4–5:30"
taskdateestimateactivity1 Tell
Type or speak your week in plain words. An AI model pulls out the tasks, dates, estimates and activities.
Try it in the planner → -
Chemistry testExamFri 13 Novno estimate, Tempo will predict? "Next Friday" can mean two dates. Is Fri 13 Nov right?
2 Check
Nothing is added until you confirm. Guesses are marked in yellow, and missing details turn into questions instead of made-up values.
-
Re-planned. What changed:
- Physics lab report added today, 1h
- History presentation part 2 moved to Sat
3 Adapt
Log how long a session really took, change an exam date, or tap "low energy". Tempo re-plans straight away and tells you what moved.
See a live plan →
03 · The AI
Three jobs, and the right
kind of AI for each.
Language model
Understanding what you type
Pick the engine at the top of the planner: Local AI (Qwen 2.5 3B through Ollama, so the text never leaves the laptop) or Groq (gpt-oss-120b in the cloud). Both reply in a fixed JSON schema.
Small models are bad at date arithmetic, so the model has to quote the student's words. Tempo's tested rules then work out the dates and check every quote against what was actually typed.
Statistical model
Predicting how long things take you
m = expΣ ln ri + k · ln m0n + k
r = actual ÷ planned minutes for each logged session · m0 = your overall habit · k = 3
With no data Tempo uses typical times. Each logged session pulls the estimate toward reality, and the range narrows as it learns.
Planning algorithm
Building a plan that never breaks the rules
Earliest-deadline-first with a slack test, spaced exam prep, a safety buffer before deadlines, subject switching, and a sleep window it never crosses.
When a week can't fit, it tries each option and only offers the ones that really work.
EngineRuns onYour text goes toSpeed
Local AIQwen 2.5 3B · OllamaThis laptop's CPUNowhereSlow: 30–90 s to read a week, ~20 s per coach answer (measured on our laptop CPU)
Groqgpt-oss-120b · free tierGroq's cloudGroq (Tempo stores nothing)Fast: ~3 s to read a week, <1 s per coach answer (measured)
Offline rulesTempo's own parserYour browserNowhereInstant
04 · Testing & improvement
We tested it, and it broke.
Here's one of those times.
We typed chem test next friday, english essay due monday about 2 hours, robotics tue and thu 4-5:30 into the local model on Thursday 5 November.
What the 3B model returned
Chemistry test: Wed 18 Nov"next Friday" isn't a WednesdayEnglish essay: Tue 10 NovMonday is the 9thChemistry test: about 2 hoursthat was the essay's estimateTask: "Robotics practice"it's an activity, not homework"I only have 2 hours on weekdays"the student never said this
What the student sees after Tempo's checks
- Chemistry test: Fri 13 Nov date corrected, flagged for confirmation
- English essay: Mon 9 Nov, 2h
- Chemistry test: no estimate Tempo will predict one
- Robotics: Tue & Thu 4:00–5:30pm, once, as an activity
- Study time: unchanged
The fix: every phrase the AI claims to quote must actually appear in what the student typed, and code we can test does the date maths. This exact model reply is now an automated test. Read all nine stories →
05 · Responsible design
Built for students,
not for engagement.
Private by default
A nickname, not a name. Everything is stored in the browser, the server keeps no logs of what students type, and there's a switch that never sends text online.
It won't do your homework
"Write my essay" gets a refusal and an offer to plan it instead. The same rule runs before every AI engine.
Kind when it's hard
A low-energy check-in lightens the day, and crunch weeks are flagged. Stress gets a kind reply that points to a counsellor or a trusted adult.
Honest about limits
Every estimate shows a range and a confidence level. Every session explains why it's there. A week that doesn't fit is never hidden.
You stay in charge
The student confirms every task the AI extracts, and picks the trade-off when a week is too full.
Accessible
Works with just a keyboard, plus large text, high contrast, extra spacing for dyslexia, dark mode, reduced motion, voice input and read-aloud.
Try it on the judges' scenario.
Three assignments, two exams, two activities and two hours a day are already loaded. Then change something and watch it re-plan.