WILLIAM LETTIERI

Codeman

Chat Channels:

Task: Use AI to develop a coding ‘companion’ that is fluent in every programming language. This assistant can plan out steps to solve problems, consider alternative approaches, and execute the composition of and entire application from start to finish.

Method: Harnessing the power of OpenAI’s custom GPT functionality, Roger Codeman was borne from pure prompt engineering and OpenAI’s latest model. Through a series of clear and explicit set of instructions, the custom GPT is dynamic and versatile.

Result: Released a hyper-intelligent friendly coding GPT partner that guides users through the entire process of development from inception to completion. The bot can browse the web for the latest updates to APIs, draw diagrams of complex processes, refactor code to meet a user’s specifications, and strategize and execute a fully-functional computer application in virtually any language.

Above and Beyond: Codeman now features the ability to connect to a user’s Dropbox, navigate and download files to review before responding, and create and delete repositories on Github. Looking forward, the next steps will be the ability to work on a Github branch and request code reviews when a push has been made. This would represent a strong step towards full autonomy and a significant increase in the productivity and capabilities of one human monitoring several Codemen.

How I Built This: Codeman Bot

Codeman Tops Coze.com

Codeman was the proud winner of Coze.com‘s AI Bot Challenge in April, 2024. 

The task was to create a productivity-centric bot on their platform. 

Soaring to over 12,000 users, 174,000 conversations, and 1,000 stars, Codeman won the award for the most popular bot on the site.

REAI

Task: Design and develop a document analysis tool that automates the extraction and classification of key data contained in lease documents, thereby enhancing the efficiency and accuracy of real property management tasks. Secondly, implement a chat bot to retrieve information from the documents to efficiently answer user questions.

Method: Utilized Python (Flask) for internal API functionality and Amazon Textract for robust PDF/OCR processing, specifically focusing on extraction of relevant data in contracts. Employed techniques like regular expressions for pattern matching and LLM-based document classification, document summarization, and extraction of critical values and legal clauses. Integrated machine learning models to generate embeddings of assembled data to facilitate semantic search capabilities for retrieval. 

Through implementation of HNSW indexing, cosine similarity and re-ranking search features, the OpenAI function-executing GPT bot rapidly searches and retrieves relevant information based on the context of the conversation and user question. Llama Indexing techniques allow searching across many documents without hallucination. 

Result: Developed a streamlined, web-based interface for property managers and real estate professionals to upload lease and amendment documents, automatically extract pertinent information for abstraction, and store it in a structured PostgreSQL database. 

The system’s embeddings-powered intelligent search functionality enables users to quickly find specific lease terms or clauses across multiple documents and generate useful reports. This  in conjunction with the assistant of the chat module significantly reduces manual review time and improves decision-making processes. This, in turn, significantly decreases administrative and derivative costs associated with human personnel. 

  

Taxman

Task: Generate a powerful GPT prompt for a character and knowledge akin to a Certified Public Accountant (CPA) knowledgeable of federal regulations and tax laws in all 50 states.

MethodHarold  Taxman is engineered to be a virtual tax advisor with the ability to make estimated tax calculations based on you or your business’s criteria. He services both simple and the most complex tax scenarios and questions.

Above and Beyond: This GPT could easily be straight out of the future of tax preparation.

Chat Channels:

Market Prophet

Task:  Develop an integrated web-based toolset designed specifically for hedge funds, providing comprehensive insights into financial market trends, with a focus on the S&P 500.

Method:  Implemented advanced data visualization techniques using Plotly charts, which included detailed volume profiles, put-to-call volume ratios, and historical analysis of seasonal patterns. Special attention was given to differentiate between election-centric and non-election-centric market cycles, ensuring a broad range of analytical tools.

Result:  Leveraged the flexibility and efficiency of WordPress for rapid prototyping and its robust built-in features, optimizing the site for user experience. Integrated JavaScript effectively to enable dynamic, real-time updates of charts without the need for page refreshes, enhancing the toolset’s responsiveness and utility for real-time market analysis.

Technical Indicator Charts

Seasonality Charts

Financial Informer

Task: Using chatbot AI techniques, program a financial news and economic event informer that can keep the user up-to-date on the latest news from financial markets and keep them abreast to upcoming economic events set to instill an impact.

Method: Coze provides the framework to harness GPT power from OpenAI for this chat bot. The bot can schedule daily news and event updates based on user choice. This bot has the ability to chat and communicate through Discord, Slack, and Telegram alike.

Above and Beyond: Aside from convenient and fascinating functionality, this the Financial Informer has penchant for conspiracies in financial markets, and perhaps rightfully so. A bias towards the belief that the system is rigged adds a much-needed dash of spice to the conversation.

Chat Channels:

Technical Indicators

Task: Construct advanced Trading View scripts, integrating the MEJT trading system into a comprehensive visual and alert-based tool. This project aimed to assist traders in identifying optimum trading opportunities in real-time, with a particular focus on clear visual representations of the system’s indicators.

Method: Utilized the scripting capabilities of Trading View’s Pine Script to translate the MEJT system into a set of visual tools and alerts. The script includes a customizable dashboard that displays key indicators such as trend lines, support and resistance levels, and potential trade setups based on MEJT principles. Special attention was given to ensuring the script’s real-time responsiveness and accuracy in different market conditions.

Result: Achieved a highly interactive and user-friendly interface, enabling traders to visually track and analyze market trends in line with the MEJT system. The integration of real-time alerts ensures timely decision-making, enhancing the tool’s practicality for day traders and short-term investors. The script’s adaptability and integration with the Trading View platform have made it a valuable tool for traders seeking to leverage the MEJT system for effective market analysis and trading strategies.

Technical Indicators

Newsman

Task: Utilize the power AI to develop a news ‘companion’ that can provide a custom-tailored news feed based on user interests.

Method: OpenAI provides the LLM base for this custom GPT’s interactivity. This super intelligent entity has several tools to access multiple News source, determine their legitimacy, and provide the latest updates based on the user’s expressed interests.

Above and Beyond: Scheduled updates provide users with a daily ‘alert’ via Discord chat software so they can be sent a message on a daily basis with up-to-date information that is relevant to them.

Chat Channels:

Chat Stream Fade

Task:  Produce a fade-in text effect that dynamically displays streaming Large Language Model (LLM) chat responses with a visually appealing entrance.
[ as seen on Perplexity.ai ]

Method:  Employed NextJS for robust frontend integration with OpenAI’s API, enhanced with TailwindCSS for sleek styling. Developed a customized fading effect to seamlessly blend incoming content.

Result:  Implementation maintained the original API response speed with negligible load cost. Entrance effect significantly enhanced user experience, creating a perception of hyper-responsiveness and a visually smooth interface, as evidenced by user feedback and engagement metrics.

Demo