Founding Cohort Demo
Dr. Vivek Kumar¶
Founding Cohort
NewsAware
A domain-adaptive, knowledge-aware newsroom agent for analyzing news and text for sentiment, credibility, and factual reliability.
Project Snapshot¶
Type
Newsroom Agent
A credibility-aware analysis system for assessing sentiment, evidence, and factual reliability in news-like claims.
Workflow
Analyze -> Check -> Refine
A simple agentic loop that extracts key claims, evaluates support, and revises credibility judgments through explicit reasoning.
What It Does¶
NewsAware is built for a problem that matters in journalism and public information: language models can produce unsupported claims, hallucinated facts, and exaggerated framing.
The live demo presents a workflow that evaluates text against three core questions:
- what is the sentiment?
- what is the central claim?
- is there enough evidence to trust it?
Core Workflow¶
The published project describes a clear agent path:
- input
- analyze
- check evidence
- refine
In practice, that means the system:
- extracts the key claim from a piece of text
- checks whether supporting evidence is actually present
- lowers or raises credibility based on the reasoning it can justify
Why Vivek Built This¶
NewsAware sits at an important intersection: AI systems, trust, and mental-health-adjacent information environments.
For a research scientist working across AI and mental health, credibility is not an abstract metric. It shapes how people interpret risk, uncertainty, and public claims. This project turns that concern into a concrete agent workflow.
Example Use Case¶
The live demo shows the system handling claims that use vague sourcing, overstatement, or conspiracy framing.
Instead of simply classifying text, NewsAware explains why a claim is weak:
- serious conclusions without verifiable sources
- vague attribution such as "reports" or "studies show"
- no cited data, institution, or study
- framing that implies suppression without evidence
Future Direction¶
The current site points toward several next steps:
- knowledge graph integration
- retrieval-based evidence checking
- multi-step evaluation agents
These are good examples of Builder-track thinking: start with a simple working loop, then extend it toward stronger evidence systems and broader orchestration.
Why It Matters¶
This is a strong founding cohort example because it turns a high-level concern about misinformation into a practical agent pattern. The project is narrow enough to be testable, but important enough to matter outside the demo itself.
It shows how SICIC projects can become applied research artifacts, not just technical exercises.
Visit the Demo¶
Closing Thought¶
NewsAware shows what a focused AI system can look like when the goal is not more content, but more trustworthy interpretation.