We write custom, robust and productionized software to solve specific use cases in the financial, industrial, energy, and technology sectors.
Business Goals:
Identify "Meme" stocks frequently mentioned in a positive light on social media platforms.
Score companies based on ticker frequency and sentiment analysis.
Identify metrics predicting stock volume and price movements.
Data Sets & Models:
Data Sources: Historical data scraped from online platforms like Reddit and YouTube, with audio transcribed to text for analysis.
Custom Finance Sentiment Modeling:
.webp)
Utilized BERT Model trained on financial corpus for sentiment analysis.
Customized phrases for each site and analysis of emojis to capture nuanced sentiments.
Intelligent Natural Language Processing (NLP) applied for accurate ticker/company extraction.
Modeling:
Several models, including neural networks, are trained to predict stock volume and price based on sentiment scores.
Outputs & Benefits:
Stock Rating Calculation:
Overall stock rating is determined by combining the Ticker Mention Frequency Score and Sentiment Score.
Portfolio Formation and Trading:
Actual portfolios are formed and traded based on the ratings derived from sentiment analysis.
Portfolios sold to retail investors seeking exposure to "Meme" stocks.
Production Environment:
Automated scraping and calculations run multiple times per day in a production environment to ensure real-time insights.
Results:
The implementation of sentiment analysis for "Meme" stock rating enabled hedge funds to identify stocks with high social media visibility and positive sentiment.
Actual portfolios formed based on sentiment analysis ratings yielded favorable returns, attracting retail investors seeking exposure to trending stocks.
The production environment running automated scraping and calculations multiple times per day ensured timely insights for trading decisions.
Conclusion:
Austin Ai empowered a hedge fund to effectively identify and assess “Meme” stocks using scaping and NLP on social media sites. Our approach broadened the fund’s library of alpha signals in a unique way, allowing them to make better trading decisions which increased return and reduced risk for its customers. The system was fully productionized in an institutional quality manner.
The combination of multiple tools like:
can be far more powerful than any one technology alone.
Succinctly www.succinctly.io is FREE software from us which allows companies to upload internal & external documents and have ChatGPT answer questions about them in a natural language format.
It is offered for FREE as an incentive to build a relationship and to bid for customization work.
Allow employees to ask questions about HR documents / polices, training manuals, product documentation, etc.
Summarize any documents like news articles, websites, public company filings, like 10-Qs, or research papers.
The more hands-on tools like Rasa & LangChain require heavy customization:
Must be fed appropriate lists of entities, topics & patterns.
Must be retrained (Rasa's neural network).
Must almost always be linked to the client's internal systems.
Outputs & Benefits:
Reduces a 2-hour manual process to under 2 minutes.
Saves multiple $MM per year.
Business Goals:
Automatically read construction blueprints from contractors.
Figure out what plumbing parts to order.
Automatically submit the list of parts into the order management system
Data Sets & Models:
Blueprints from the client's clients.
OCR.
NLP on the OCR results.
Business Goals:
Forecast how dirty panels on solar farms will get based on surrounding conditions
Compute optimal time to wash the panels (which is very expensive) versus electricity lost from soiling
Data Sets and Models:
Particulate matter trends from EPA
Weather history from NOAA interpolated over a grid over the entire USA
Output and Benefits:
Makes $MMs per year per solar farm in reduced washing costs and increased power generation
Better for environment in terms of water usage
Business Goals:
Evaluate the credit & payment history of
general contractors, property owners, and hiring parties.
Extend the appropriate amount of credit in project financing deals.
Anticipate & reduce write-offs.
Augments traditional credit reporting tools/companies.
.webp)
Company A
Payment Risk Score
669
6431 jobs in the last 6 months per information available.
Job Growth
27%
Industry Average: 5%.
Payment Speed
66
days
Industry Average: 88 days.
Dispute index
2%
Industry Average: 5%.
★ 4/5
17 Ratings
Social Sentiment
👍Positive
Common Job Types
Residential
Annual Sales
Over $35 B
Employees
5000-9999
123 Sunset Blvd, Hilltown, CA 99922
Data Sets & Models:
Many large, disparate data sets:
Large databases of construction project, lien, invoice, company, etc.
Graph database of network effects / relationships
User reviews & ratings.
Extensive decision tree computes hundreds of clean metrics.
NLP sentiment analysis on customer reviews.
Outputs & Benefits:
Risk reports sold to third parties as a data product:
Potential six-figure additional revenue.
Internal credit rating aids in internal credit decisions:
Reduces default rates by 30-40%.
Construction industry statistics for marketing, white papers, etc:
Bolsters firm reputation.
Business Goals:
Analyze browsing of retail web users and predict when they are about to purchase something
Serve advertisements / discounts at the optimal time
Data Sets and Models:
Time series of domains, search terms, timestamps, geographies of users
Intelligent transformation of original inputs
Several features using large language models and/or NLP
Output and Benefits:
Predicts 7 out of 10 user purchases.
Predicts Amazon category the user is interested in.
Optimizes advertising spend and timing.
%201.webp)
Outputs & Benefits:
Composite investment indicator.
Visualization by zip code on an interactive map.
Identification of outliers (under- or over-valued locations)
Business Goals:
Grade physical locations on investment attractiveness.
Identify trends in demographics & other time series.
Relate property values to explanatory variables & their rates of change.
Data Sets & Models:
U.S. Census from 2012 (TB of data).
Zillow home price estimates.
Yelp reviews.
School reviews.
Business Goals:
Anticipate equipment failure.
Increase predictive maintenance.
Reduce reactive service calls by 30%.
Outputs & Benefits:
Variables & patterns most related to future failure;
Provides engineering insight into failure points.
Probability of failure within various time periods.
Model statistics like precision, recall, false positive rates, etc.
Anticipation & reduction of service calls:
Calls reduced by 30+%.
Costs reduced by 20+%.
Many on-demand calls transformed into anticipatory ones.
Data Sets & Models:
Log data from equipment.
Sensor readings.
Error, warning, and status codes.
Failure flag.
Machine ID#'s and diagram of manufacturing process.
Both random forests & neural network models trained on data.
