Candour

simultaneous working of two engines.
5 step agent work flow
data pipeline
asking query as an actor, submitting claimed credits and collaborations
Evidence analysis of the query asked as an actor
displaying the feature of printing/saving report by the user.

Inspiration

We always wanted to know how the information flowed in the cinema industry. And after research we understood there's no reliable resource to analyze their data! Thus we came up with Candour. Candour analyzes your opportunity to give you trustworthy insights, without guess work.

What it does

Candour takes your role, your opportunity, if any claimed credits by the person you wish to check and their collaborations (credits and collaboration being optional inputs), then runs a query, finds evidence based on your claim and provides you a printable analysis, which speaks fact. Every query passed is answered with respect to your role selected.

How we built it

We used 3 databases: TMDB, Wikidata and IMDB in ClickHouse. Our agent runs on Google ADK with Gemini, and it queries ClickHouse live through the official mcp-clickhouse MCP server, so every answer comes from a real-time query, not a precomputed lookup. Backend is FastAPI, frontend is React + Tailwind.

One of the harder pieces was identity resolution. Since the same person can show up differently across TMDB, Wikidata and IMDB, we built a tiered matching system that disambiguates who's actually who before running any analysis on them.

Challenges we ran into

Free tier of API wasn't enough for testing. TMDB API took 25 hours to load data into ClickHouse. Reconciling three data sources was tricky, since the same person often has inconsistent identifiers or credits across TMDB, Wikidata and IMDB. We had to build matching logic to disambiguate who's who before trusting any of the data. We wanted every answer to be honest rather than guessed, so when a data source was incomplete, our system had to know to say "we don't know" instead of making something up.

Accomplishments that we're proud of

Our major accomplishment is that we built a whole fullstack project in a tight deadline, overcoming the challenges in between and deploying it with only free resources. We're also proud that our agent queries ClickHouse live through the official MCP server, so every insight comes from a real query and not a precomputed answer. And we stayed true to our own promise of no guesswork — when our data coverage was incomplete, Candour says so honestly instead of faking a number.

What we learned

It was quite a journey! We learnt a lot about ClickHouse OLAP, which saved us a lot of time with managing queries. We also understood some core concepts of full stack. We also learnt a lot about agent design, like how to prompt and orchestrate an agent that actually queries a live database instead of just answering from what it already knows. We had to teach it to know when it doesn't have enough data, and say so, instead of guessing an answer just to sound confident.

What's next for Candour

After doing the product testing with National School of Drama's students, we've realised that Candour is a genuine help to the community. Next features on the list for Candour are:

Showing info-graphics.
Finding more reliable sources of data for smoother operation.
Adding more roles beyond actor, writer, indie crew and investor, so more people in the industry can check who they're working with.
Hardening it for real production use, since right now it's built for a hackathon demo, not everyday traffic.