Dainik Samachar

How Dainik Samachar works

From a noisy internet to research you can trust.

Dainik Samachar is a research-oriented news intelligence system for financial, economic, and market topics. It searches the open web, keeps the sources, and writes briefings, chat answers, and stories only from that evidence. Discovery, conversation, and publishing are different paths through the same record.

  1. Internet
  2. Research
  3. Evidence
  4. Understanding

The problem

The internet has information. The problem is finding what matters.

Search returns fragments. Chat forgets what it already read. Every extra search and scrape costs money. A briefing is only useful when the sources are still attached to it.

  • Headlines
  • Videos
  • Threads
  • Articles
  • Sources
  • Documents
  • Claims

Signal

From noise to signal

A wide field of sources, narrowed to something you can read.

News discovery and chat research both pass through search, selection, scraping, and synthesis. What remains is stored separately from the words written on top of it.

Internet

  • Articles
  • Videos
  • Search results
  • Pages

Dainik Samachar

  1. Search
  2. Select
  3. Deduplicate
  4. Extract
  5. Evidence
  6. Synthesize

Research

  • Story
  • Sources
  • Context

What you can do

Three research paths. One stored record.

A briefing discovers many stories. Chat answers one question. Asking for a story turns that research into a document you own. Every path writes sources before it writes prose.

  1. 01

    News briefing

    A date, place, and set of topics become a multi-story digest with sources still attached.

  2. 02

    Community stories

    Published stories sit on a public feed. Readers vote, bookmark, and open a deep dive.

  3. 03

    Research chat

    A finance or economics question is researched, answered, and remembered on the session.

  4. 04

    Story deep dive

    A published story opens a chat that starts from its stored sources, not a blank search.

  5. 05

    Chat to story

    Asking for a story creates a private draft from research already gathered, then you can publish it.

  6. 06

    Location-aware search

    Autocomplete and reverse geocoding bias local briefings without calling Maps on every keystroke.

How Dainik Samachar works

One system. Several research paths.

A news request becomes many stories. A chat message becomes an answer—or, on a general chat, a draft story. A deep dive starts from a story you can already open, then continues as chat.

  1. 01Reader
  2. 02Next.js + Clerk
  3. 03Inngest
  4. 04News · Chat · Chat story pipelines
  5. 05SerpAPI · Firecrawl · YouTube · OpenAI
  6. 06PostgreSQL + pgvector

News research

A briefing is a planned search, not a single query.

The news pipeline runs as one durable Inngest job. YouTube can support a story; it cannot replace a primary article.

  1. 01

    Discover

    A news request: date, place, scope, topics.

  2. 02

    Search

    Planned queries across Google News, web news, and YouTube.

  3. 03

    Collect

    Normalized hits, plus AI Overview follow-up searches.

  4. 04

    Select

    An article selector picks what is worth reading.

  5. 05

    Scrape

    Firecrawl turns chosen URLs into article text.

  6. 06

    Clean

    A cleaner strips navigation, ads, and junk pages.

  7. 07

    Synthesize

    Stories are written only from that evidence.

  8. 08

    Story

    NewsStory and NewsSource rows, then a notification.

Conversational research

Ask a question. The system researches it.

Chat is narrower than a briefing. It reuses this session's research when similarity is high enough, and it searches only when the determiner says the record is not enough.

In a session

“What changed in this story since yesterday?”

Two of the four sources already in this session cover the change. One new report was added after a targeted search. Here is what moved, and what did not.

  • reuters.com
  • ft.com
  • youtube.com/watch…
The first turn of a deep dive starts from the story and its stored sources instead of a fresh search plan. Every turn after that runs this same process. Conversation text is context for the model; it is never cited as a source.

  1. Question

    A message in a research session.

  2. Understand

    Guardrails keep it to markets and economics; a query enhancer and a determiner plan the work.

  3. Check memory

    Similar session research from pgvector, when this session already has some.

  4. Search if needed

    SerpAPI, YouTube, and AI Overview follow-ups only when the determiner asks.

  5. Collect evidence

    Firecrawl, cleaning, saved ResearchSource rows.

  6. Answer

    A reply grounded in that evidence.

From conversation to story

A question can become a document.

On a general chat—not a story deep dive—the determiner can hand the research it already gathered to a one-story pipeline. Nothing is searched twice for the same evidence.

ChatCreate storyStory

  1. 01

    Question

    Asked in a general chat.

  2. 02

    Chat research

    The message pipeline gathers evidence.

  3. 03

    Create story

    The determiner decides a story is wanted.

  4. 04

    Story pipeline

    Reuses that evidence; searches more only for gaps.

  5. 05

    Evidence

    NewsSource rows saved beside the prose.

  6. 06

    Draft story

    Private until the owner publishes. The owner can edit copy and upload a cover photo.

The result is a NewsStory you own. It stays a private draft until you publish it. You can edit the copy and upload a cover photo. After publishing, other readers can open it, vote on it, and deep-dive it. Only you can edit it.

Evidence

The writing sits on top of stored sources.

There is no separate claims database. A briefing keeps NewsSource rows on the story. Chat keeps ResearchSource rows on the session. Synthesis reads those texts—not the transcript of the conversation.

  1. Documents

    Article text · YouTube transcript · Session research

  2. Sources

    NewsSource · ResearchSource

  3. Story

    Story

Research memory

Earlier research can answer the next question.

When a chat source is saved, a short description is embedded with text-embedding-3-small and stored in pgvector. Later messages in the same session retrieve the closest rows (top 8, similarity ≥ 0.72) before spending another search.

● this question, among earlier session research

  1. Previous research
  2. Embedding
  3. Vector space
  4. Similarity search
  5. Relevant research
  6. Current question

Retrieval is scoped to the current session. Chat messages are also embedded in the background, but that index is not read by any pipeline yet; answers are written from research sources.

Engineering

Built as a research system, not a chatbot wrapper.

Long research does not live inside one HTTP request. Inngest runs the jobs and retries them. Postgres keeps the sources. Models write only after the text is in hand. The repository README is the full technical specification.

  1. 01ApplicationNext.js · TypeScript · React
  2. 02OrchestrationInngest
  3. 03ResearchSerpAPI · Firecrawl · YouTube · OpenAI
  4. 04StoragePostgreSQL · Prisma · pgvector
  5. 05IdentityClerk

What is kept

From the open web to something a person can finish.

  1. 01Raw web
  2. 02Search results
  3. 03Scraped pages
  4. 04Cleaned text
  5. 05Stored sources
  6. 06Synthesis
  7. 07Story or answer

Read less noise. Understand more.

Start with a briefing, or open a research chat and follow one question until the sources hold.

Explore Dainik Samachar