govLens

• 

• 

• 

GovLens

Inspiration

We realized something uncomfortable.

Most of us only find out about policies that affect our lives after it is too late.

Whether it is immigration changes, housing protections, transportation rules, or education funding, we often learn about major decisions through headlines or social media. By then, the information is filtered, biased, or incomplete. And even when we hear about something, we rarely know:

• Who introduced it • When it takes effect • Whether public comment is still open • Who we can contact • What we can actually do

In other words, we are reactive when we should be informed and proactive.

That frustration inspired GovLens.

We wanted to build a system that does not just summarize government updates, but enforces truth, surfaces uncertainty, personalizes impact, and helps people take action with confidence.

What it does

GovLens is an AI powered civic intelligence platform that transforms complex government updates into verified, personalized civic alerts.

It:

• Surfaces current updates from official sources • Clearly labels who is affected • Flags urgency such as deadlines or meetings • Provides evidence quotes and source links • Allows users to ask, “What does this mean for me?” • Guides users on what steps they can take next

Every update becomes actionable. Users can start a contextual chat thread tied to a specific issue and receive grounded responses with citations. Their interactions are saved in a civic inbox so they can follow up, revisit, and track what matters to them.

Truth is the foundation. Service is the purpose.

How we built it

We designed GovLens with guardrails first.

Rather than letting AI freely generate civic information, we implemented:

• Structured output schemas to enforce consistent data • Evidence title alignment checks • Domain quality filters to block unreliable sources • Deterministic downgrade rules when verification fails • Confirmation logic that requires grounded quotes

Each update must include a source link and an evidence quote. Weak sources are downgraded. If something cannot be confirmed, it is marked as uncertain or not found. The system never fabricates civic claims.

For performance and reliability, we built versioned caching, deterministic card identifiers, strict validation layers, and capped citation logic to maintain clarity in the user interface.

We intentionally separated:

• Feed generation • Validation and ranking • Chat thread persistence • Context personalization

This allowed us to build a minimum viable product that remains stable under live demo conditions.

Challenges we ran into

Time was our biggest constraint.

Within 24 hours, we had to:

• Design a clean, intuitive feed • Build a validation engine that prevents misinformation • Implement structured AI extraction • Create a contextual chat experience • Add caching and persistence • Ensure the UI never breaks even when data is missing

Another major challenge was evidence alignment. We had to prevent scenarios where a source quote did not actually support the claim being shown. This required implementing token overlap checks, domain filtering, and deterministic ranking logic.

Balancing transparency, personalization, technical feasibility, and ethical responsibility was the central design challenge of the project.

Accomplishments that we're proud of

We are proud that GovLens does not guess.

We built a system that:

• Downgrades weak or unverifiable sources • Never confirms fallback domains like Wikipedia or Ballotpedia • Uses deterministic validation rules to enforce truth • Always returns stable, structured UI outputs • Preserves user context in a civic inbox for follow up

In a short time window, we built not just a chatbot, but a verification first civic intelligence system grounded in evidence.

What we learned

The hardest part was not building a chatbot.

The hardest part was building trustworthy AI.

Civic information carries real consequences. Inaccurate summaries or hallucinated details are unacceptable in this domain. We learned that responsible AI requires:

• Explicit validation layers • Deterministic downgrade logic • Clear uncertainty signaling • Transparent citation practices • Data minimization and ethical design

We also learned that personalization increases engagement. When updates are framed around a user’s location or priorities, civic information becomes relevant instead of overwhelming.

What's next for GovLens

Government decisions shape our daily lives.

GovLens reflects the principle of Excellence in Truth and Service, but this is only the beginning.

Next, we plan to:

• Expand beyond DC to additional jurisdictions • Improve personalization using richer user context • Add proactive alerts tied to deadlines and public comment windows • Integrate direct action tools such as verified contact routing • Continue strengthening validation logic to reduce misinformation risk

Our long term vision is to make GovLens a transparent AI layer that helps communities understand government in real time and take meaningful action with confidence.

Built With

• expo.io
• fastapi
• google-web-speech-api
• mongodb
• openai
• python
• reactnative