Sentimint

Inspiration

We were tired of stock advice that either felt like fortune cookie wisdom or required a CFA to decode. So we asked ourselves:
What if you could combine real-time financials + breaking news + LLMs, and turn that into a judgmental stock analyst who speaks human?
Boom. Sentimint was born. 🌿

What it does

Sentimint is your friendly finance whisperer. You type in a company or ticker like Netflix or NFLX, and it:

Scrapes real-time financial data 🧾
Reads the latest headlines from Yahoo Finance 📰
Generates a recommendation (Buy/Sell/Hold) via a smart LLM 🤖
Verifies itself using a second agent with multi-source news + heuristics ✅
It’s like having two financial interns — one optimistic, one paranoid.

How we built it

LangChain for multi-agent reasoning
OpenAI’s GPT-3.5 for natural language insights
Selenium + BeautifulSoup for news scraping
A hashtable + CSV magic to resolve company names to tickers
Pandas + Heuristics to flag suspicious metrics (e.g. 🚩 PE > 100)

Challenges we ran into

Real-time scraping is chaotic neutral 🌀
Finetuning prompt logic without turning our LLM into a ✨financial poet✨
Teaching our agents to double-check each other without starting a debate club
Resolving "Apple" to AAPL without accidentally choosing APPLE Hospitality REIT

Accomplishments that we're proud of

Built a working multi-agent financial decision stack
Implemented real-time frontend updates using WebSockets
Created a verifier bot that acts like a skeptical co-founder
Handled both tickers and company names like a real product

What we learned

LangChain tools are powerful (and sometimes too powerful)
Verifier bots dramatically increase trust in LLM pipelines
Adding heuristics + contradictions makes agents more robust

What’s next for Sentimint

Add multi-language support for global users 🌍
Build a dashboard view for comparing multiple stocks side-by-side 📊
Integrate real-time earnings reports + Reddit sentiment
Let the verifier call a human or generate a calendar invite when confidence < 50%