Truthy

Value Proposition

As an immigrant, when you first submit an application to the Canadian ministry of immigration (IRCC), it takes monnths, if not years, for that application to be reviewed for an initial "R10 Completeness Check" before entering the queue for processing. This is a laborious human-led and, hence, a highly time-consuming process. Truthy is a Business-to-Government (B2G) AI-powered automation application that has aimed at reducing this process from months to seconds, hence leading to faster overall processing of IRCC applications.

Prizes We're Competing For

We'd like to compete for the following prizes:

• Google Best AI for Community Impact (In-Person Only)
• Bitdeer Beyond the Prototype: Best Production-Ready AI Tool (In-Person Only)
• Moorcheh AI Best AI Application that Leverages Efficient Memory (In-Person Only) If each team is limited to selecting one prize, our prize selections are written in order of preference. That is, we most desire to compete for the 'Google Best AI for Community Impact' and least desire to compete for the 'Moorcheh AI Best AI Application that Leverages Efficient Memory'.

Services / Architecture Summary

The platform provides the following technical services:

1. Main FastAPI gateway and orchestration layer. This is the system entrypoint and calls the downstream indexer and agentic-rag services.
2. Streamlit-based internal dashboard frontend for officers, analysts, and demo workflows.
3. Policy ingestion and indexing service for program guides, checklists, form instructions, and rule updates.
4. agentic-rag: Retrieval and reasoning service that analyzes extracted application data against indexed policy knowledge.
5. Fast Redis cache system where a policy's modification date on exisiting regulations' database is frequently checked against the live version to determine if amendment to the database is required.

The platform is structured as a service-oriented monorepo:

1. streamlit provides the user interface.
2. api receives UI requests and coordinates workflows.
3. api calls indexer for policy ingestion and refresh workflows.
4. api calls agentic-rag for retrieval, reasoning, and completeness analysis.
5. Shared contracts, prompts, and documentation live at the repository level.

Challenges we ran into

There are many sources outlining the law or providing operational guiding instructions to IRCC Officers for how to perform the so-called "R10 Completeness Check" in relation to each specific application. Our challenge was to feed these data, either as system prompts or in our indexer, such that the model would not read or be instructed with conflicting information, thereby leading to inconsistent decision-making.

Accomplishments that we're proud of

In a highly time-intensive process, we managed to brainstorm the above problem statement, solution, assign tasks in between group members, implement and debug our application in only approximately half a day!

What's next for Truthy

Currently, Truthy is enabled to perform the "R10 Completeness Check" on only two IRCC programs, Temporary Resident Visas and Study Permits (from outside Canada). Following the current implementation, our team aims to expand to the 350+ distinct programs existing within the Canadian immigration system. Furthermore, we wish to increase our AI model's accuracy in scavenging the content of files and read through different file types (word, jpg, png, etc.). In tandem, we will also attempt to enhance our model's reasoning capabilities moving forward.

Built With

• agentic
• css
• docker
• html
• indexer
• javascript
• python
• qdrant
• rag
• redis
• streamlit