🚀 Inspiration

Every serious researcher — from grad students grinding out their lit reviews to tenured professors trying to track five subfields at once — knows the pain:

  • New research moves too fast
  • It's scattered across a dozen platforms
  • And if you blink, you've missed a whole trend

Between arXiv, Twitter threads, conference keynotes, and shadow releases on GitHub, there’s no single feed that captures the full signal.

We weren’t inspired by another one-shot summarizer. We were inspired by how actual researchers work:

  • Constant pivots between sources
  • Zooming in and out across timescales
  • Evaluating claims, not just regurgitating them
  • Prioritizing trust, depth, and novelty

So we built Vizier — not another autocomplete wrapper, but a system that actually understands your goals and assembles a live research team around them.

Because in 2025, Insight isn’t just another token.


🧠 What Vizier Actually Does

Vizier is a modular, agentic research engine — your personal research ops team, not just a chatbot. Whether you're building a newsletter, writing a paper, pitching an idea, or just staying on the bleeding edge, Vizier gives you:

  • 🔍 Precision-curated content from credible, multi-platform sources
  • 🧱 Structured, editable reports tailored to your research priorities
  • 🧠 Control over what gets emphasized, where deeper sourcing is needed, and how frequent updates should be

It’s built for:

  • 🧑‍🔬 Researchers and technical professionals who need rigorous updates on specialized domains
  • 🎓 Students and professors tracking rapid fields like GenAI, climate science, or synthetic bio
  • 📣 Content creators and analysts writing newsletters, reports, or breakdowns on bleeding-edge developments

And it doesn't stop after one search. Once you've locked in a great output, you can:

  • 📅 Schedule that research plan to auto-run daily, weekly, or monthly — get a fresh update, no extra effort
  • 🔁 Revisit past reports, tweak scopes, swap source weights, or layer in new domains

🔁 What Makes Vizier Agentic?

This isn’t just “use LangChain and call it a day.” Vizier’s agents think for themselves.

  • 🧭 Router v0_4 analyzes your refined query and decides:

    • How many agents to spawn
    • Which domains get which budget
    • Which model contexts are needed (Twitter thread agents ≠ PDF interpreters)
  • 🔎 Web and Twitter Search Agents don’t just follow rules — they evaluate:

    • How noisy a domain is
    • Whether depth is sufficient
    • If a second-level validation or domain rerun is required
  • 🧪 Source Review Agents actively rerank or prune content if trust scores fall short, or if a source looks promising but under-explained. They’re built to push quality higher — not just pass things through.

Every Vizier agent has decision power — they can go deeper, re-query, or backtrack if a lead is weak. They operate under flexible planning constraints, just like human researchers.


🛠 How We Built It (No Bullshit Edition)

  • 🧠 Query Refiner
    Builds a multi-component research plan with you — grounded in:

    • Role (e.g. systems researcher, policy analyst, VC)
    • Goals (e.g. “deep tech convergence tracking”)
    • Update frequency (e.g. daily pulse, monthly digest)
  • 🧭 Router v0_4
    Not a switchboard. A planner. It:

    • Maps query scope to modality
    • Assigns sourcing budgets
    • Splits loads across independent agents with context-aware prompts
  • 🌍 Agents That Decide

    • Web & Twitter agents determine their own search depth
    • Can re-run if a thread lacks references or a blog post is all hype
    • Know when to cross-link sources or bring in institutional citations
  • ✍️ Writer Agent
    Doesn’t dump. It synthesizes:

    • Modular paragraphs per component
    • Auto-queries clarification if details feel fuzzy or under-sourced
    • Lets you re-rank or flag parts for improvement
  • Live Agent Graph UI
    SSE-driven, real-time frontend that shows the thinking — you can trace every decision, branch, retry, and rerank in real-time. This is visible cognition, not a black box.


🔧 Challenges We Solved

  • 🧬 Multi-agent coordination with branching workflows & retries
  • 🔁 Live SSE sync between asynchronous agent outputs and frontend UI
  • 🧠 True agentic control flow: Where to go deeper, when to bail, how to reprioritize
  • 🐦 Navigating Twitter’s API hell: Index instability, partial results, thread collapsing — all solved with multi-layered backup scrapers, re-verification agents, and de-noising heuristics

🌟 Accomplishments We’re Proud Of

  • Real agents that take real actions — they don’t just forward results, they analyze and rerun as needed
  • 🧠 Router with context-driven source allocation, not hardcoded domain splits
  • 🧮 Multimodal synthesis across whitepapers, blog posts, podcasts, preprints, Twitter threads, and more
  • 🔁 Feedback on both ends — not just user ↔ system, but agents ↔ other agents (e.g. writing agents asking for clarification from sourcing agents)
  • Interactive report scheduler — see something good? Auto-run it next week, no retyping needed

📣 Why Vizier Actually Matters

If you’re a researcher, analyst, professor, or creator — Vizier helps you do the real work better and faster, without sacrificing rigor.

  • 🧠 Tired of missing Twitter alpha from top researchers like Karpathy, Srivastava, or Andreessen? Vizier gets it before the news cycle does.
  • ✍️ Publishing a newsletter on Substack or Beehiiv? Vizier lets you build a content pipeline that you can iterate, edit, and share easily.
  • 🎧 Want to convert a weekly report into a NotebookLM podcast, or generate highlights for Twitter and LinkedIn? That's on the roadmap.
  • 💰 Want Stripe-powered revenue sharing on recurring reports or private briefings? We’re building toward that, too.

Vizier isn’t just a tool. It’s a workspace.
One that grows with your workflow, so you can eventually do everything from query to publication in-platform.

No more tabs.
No more copy-pasting between Notion, ChatGPT, Zotero, and Twitter.
Just signal, from chaos to clarity.

Built With

Share this project:

Updates