TasteCompass

Your tastes, scored on Qloo's UK-wide heatmap: the best-fitting student cities light up, ready to match with real programmes.

Inspiration

Choosing a UK master's isn't only about rankings. You spend a year living in a city, and most advice ignores whether you'll actually like it there. I'm applying to UK programmes myself, and the hardest part wasn't finding courses. It was finding a city that fits how I live, and getting deadlines and fees I could trust. General chatbots answer confidently from memory, often with last year's dates.

What it does

TasteCompass is an agent that matches your taste to UK cities and master's programmes, then plans your application.

Say what's decided: nothing yet, the country, a city, or specific programmes. Whatever you fix becomes a hard filter.
Share your tastes: artists, films, books, food and hobbies.
Get ranked programmes: each scored on taste (Qloo), academic fit, budget and English requirements, with weights you control.
Get a dated plan: when to book IELTS, ask referees and submit, worked back from the next open application round.

If you've already chosen programmes, it checks them and flags better-fitting alternatives. For example, it looked at my own UCL and KCL picks and surfaced two Bristol programmes that fit my taste better.

How Qloo powers it

Without Qloo this would be another search box. Qloo is what makes "where would I be happy?" measurable:

Understanding taste: /search (restricted to works and people) and /v2/tags turn each interest into a Qloo entity or tag. Radiohead becomes an artist entity; hiking becomes an activity tag.
Comparing cities on one scale: a single UK-wide /v2/insights heatmap scores the 24 major student cities on the same scale, measured within 10 km of each city centre. (We found per-city place affinities can't be compared across cities, because each city is ranked on its own curve.)
Showing what you'd love there: Qloo place recommendations, filtered to the city and cleaned of national chains, become a "You'd like in Bristol" list and a first-week plan.

Every number has a source

13 programmes at Oxford, Cambridge, Imperial, LSE, UCL, KCL, Edinburgh, Manchester, Bristol, Glasgow and Leeds were checked by hand on official course pages for 2027 entry, each with its link and the date checked.
110 universities in 63 UK cities are listed with their official sites.
Anything a course page doesn't state is shown as unverified and scored neutral, never guessed.

How it compares to a general LLM

We asked GPT-6 (web search on) 10 factual questions about 2027 entry: deadlines, fees and IELTS for UCL, KCL, Oxford, Imperial, LSE, Edinburgh and Glasgow. It got 8 right. Both misses were cases where the official page had already published the figure, but the model said it was "not yet confirmed" and fell back on a third-party site or its own estimate. It gave a UCL visa-applicant deadline 5 days too late (27 Mar instead of 22 Mar 2027), which would mean missing the deadline, and a KCL fee £1,316 too low. It also labelled a correct KCL deadline as unconfirmed. TasteCompass answers all 10 from official pages, with sources. (Small spot check: 10 questions, one run, chosen from programmes in our dataset.)

How we built it

Agent: Python. It reads your scope, picks which tools to call, applies hard filters, scores, plans, and keeps a trace of every step, shown in the UI as "What the agent did".
Qloo Insights API: search, tags, the UK heatmap and place insights, with retries and a disk cache.
Optional LLM layer: OpenAI explains results and answers what-if questions ("what if my budget were £3,000 higher?") by rerunning the real scoring. It never supplies facts.
Frontend: a homepage plus an agent UI in plain HTML/JS, deployed on Render.

Challenges

The first city ranking was noise: every city scored about 0.80 because place affinities are ranked within each city. Switching to one national heatmap fixed it.
Across 63 towns, the heatmap mostly tracked how busy an area was (correlation 0.93 with popularity), so Bolton outranked Bristol for an indie/street-art profile. We now rank taste across the 24 major student cities and list universities everywhere else.
Free-text interests: "Hiking" first matched a trail in Rio and "Harry Potter" a dish. We now resolve activities to tags and works to entities.