NexBrief AI
A working news product, not a demo. NexBrief AI presents the Left, Centre and Right reading of a story side by side, rates the credibility of every source it shows, and includes a fake-news detector that scores a claim and tells you how to check it yourself. Live at newz.srbros.in with accounts, a trial and paid tiers.
In progress: Actively being built right now.
Opens newz.srbros.in in a new tab.
Placeholder artwork. Project photography, screenshots and video are pending — see the asset checklist in the repository.
Problem
Two people can read the news every day and end up with incompatible pictures of the same event. The cause is rarely invented facts — it is selection and framing. Which details lead, which are buried, which words are chosen. A reader seeing one outlet has no way to tell what was left out, and reading across the spectrum manually costs time nobody has.
- Framing differences are invisible when you only read one source
- Comparing outlets by hand is slow and most people will not do it
- Treating a summary as neutral does not make it neutral
Solution
Four tools built around one idea — that a reader is better served by seeing the disagreement than by being handed a verdict.
- Compare — enter any topic and get Left, Centre and Right perspectives side by side, along with what each one leaves out
- Daily News — a live feed from multiple newsrooms, refreshed every 60 seconds, with a credibility rating on every story
- Detect — paste a headline, claim or URL and get a credibility score, the red flags behind it, and steps to verify it yourself
- Ask anything — a general question box that is not restricted to news; science, history and economics all work
Source credibility
The part that does the most work is the least glamorous. Every outlet in the feed carries a credibility rating — High, Mostly High, Mixed or Low — synthesised from independent evaluators including Media Bias/Fact Check, Ad Fontes Media and NewsGuard, plus a stated bias position. Reuters and AP News sit at High · 95, Centre. The Hindu appears as High · 85, Centre / Slight Left.
- Ratings measure factual reporting, sourcing standards, editorial independence and correction practices
- Bias placement is shown explicitly rather than implied
- Each story carries a one-line characterisation of the outlet publishing it
- Every card links out to the original article so the reader can check it
Architecture
There is no traditional server. The front end is a React 18 and TypeScript application built with Vite, styled with Tailwind and shadcn/ui on Radix primitives, routed with React Router and fetching through TanStack Query. Everything behind it runs as serverless Deno functions on Supabase.
- chat — streaming AI with four modes: answer, compare, fact-check and detect. Accepts images as well as text
- news — merges 14 RSS feeds including Reuters, AP, BBC, The Times of India and The Hindu. No AI involved: the feed is real articles, unmodified
- translate — batch translation, chunked and parallelised for speed
- Google sign-in through Supabase Auth using PKCE, the correct flow for a public client
- Postgres with two tables and row-level security, so access rules live in the database rather than in application code
- A signup trigger creates the profile and starts the 14-day trial automatically — no app-side bookkeeping to get wrong
Decisions worth defending
A few choices in this build are the kind that only look obvious afterwards.
- Gemini is called through its OpenAI-compatible endpoint, so changing model provider is a URL change rather than a rewrite
- Keeping AI out of the news function entirely means the feed cannot hallucinate a headline — credibility ratings attach to articles that genuinely exist
- Row-level security means a bug in the front end cannot leak another user's data
- Serverless throughout: nothing to patch, nothing to keep running at 3am
The hard part
The difficult questions here are editorial, not technical. What counts as the Centre, and centre of what — India's political spectrum or America's? How do you represent a position accurately without adopting it or caricaturing it? When two credible outlets disagree on a fact rather than a framing, what does the neutral summary say? A rating system that places outlets on a spectrum is itself making a claim, which is why NexBrief cites external evaluators rather than inventing its own scores.
Shipped as a product
This is where the project stops being a demo. It has real accounts, real billing tiers and a real trial — the unglamorous work most side projects skip.
- Google OAuth sign-in with a 14-day trial, no card required
- Basic at ₹250/month and Pro at ₹340/month, with annual pricing
- Language selector and a light/dark theme toggle
- Instructions and FAQ written for users who have never seen the product
Status
Live at newz.srbros.in and in active development. Compare, Daily News, Detect and the ask-anything box all work today. Saved research is announced in-product as coming soon, and subscription activation is currently handled manually rather than through an automated payment flow.
Possible future improvements
Status: ConceptNone of the items below have been built. They are recorded here as direction, not capability.
- Grounding Compare in the RSS layer, so every perspective cites an article the system actually retrieved
- Saved research — currently marked as coming soon in the product
- Automated subscription activation, replacing the manual admin step
- A public demo view so the work can be seen without signing in
- Wider source coverage, including regional and non-English outlets
- Tracking how a running story's coverage shifts over time