CYPHER

No time to read the Annual Report? Listen to the interrogation. Automated financial due diligence in a podcast format.

Comment

Inspiration

To be honest, we didn't start the 24-hour hackathon with this idea. We sat down with a blank screen and a vague goal: "Make financial research easier."

Our first thought was a standard "Research Paper Summarizer." But as we prototyped, we realized a hard truth: Summaries of 100-page reports are still boring to read. We asked ourselves: Why do people prefer financial podcasts over PDFs? The answer is narrative tension.

We pivoted immediately from a text tool to CYPHER.

What it does & How we built it

We wanted to turn due diligence into a "True interesting" style debate—where the Company Report is the suspect, and the News is the detective. We also generate analysis report by using Anthropic and also Apify to find, download and parse the latest Annual Report(PDF) and real time news for any ticker.

  • The Brain (Claude 3.5 Sonnet): We prompt the AI to take sides. One agent acts as the "Bull" (quoting the report), and the other as the "Bear" (using news to contradict the report).
  • The Voice: We use ElevenLabs to assign distinct personalities to the speakers, turning the JSON script into a radio-quality broadcast.
  • The Database: We use Convex for the data storage.

Challenges we ran into

-Data Overload: Annual reports exceed AI context limits. We built logic to surgically extract only the critical "Risk" and "Financial" sections.

Accomplishments that we're proud of

  • Built a fully autonomous loop in 24 hours: We successfully chained 3 separate AI agents (Search, Reasoning, Voice) to run without human intervention.
  • The "Audio-to-UI" Sync: We didn't just play an MP3; we engineered a way to make the dashboard react to the audio in real-time, creating a cinematic experience in a standard web browser.

What we learned

-Narrative > Raw Data: People don't want to read reports; they want to hear a story. The "Bull vs. Bear" format proved that framing data as a conflict makes it instantly engaging.

  • Agents need "Personality": Giving the AI specific roles (the "Ruthless Skeptic" vs. the "Corporate Optimist") stopped the output from sounding like a generic ChatGPT summary.

What's next for CYPHER

  • True Video Generation: Replacing our frontend "sync hack" with actual AI-generated video avatars (using HeyGen or similar APIs) for a full visual broadcast.
  • Mobile App: Moving this from a web dashboard to a morning commute app.
  • Beyond Stocks: Expanding the "Debate Engine" to other verticals like Political Bills, Crypto Whitepapers, and Legal Contracts.

Built With

Share this project:

Updates