.png)
Our deployed Solutions and case studies
Explore our deployed solutions and case studies to gain insights into our successful projects and discover actionable strategies for your own business.
AI Course builder
Technologies Used
GPT3, Langchain, Stripe, FastAPI, Python, Mysql
Client Objective
The client objective was to build a chatgpt plugin, in which the user will login and have OAUTH authentication. Using the plugin, for any given topic let user get the curriculum which will be a document consisting of 14-25 pages long which will have table of content, Paragraph, Activity Matrix and MCQ for each topic. For the plugin, we had to develp a payment gateway too where the user can buy the token from stripe.
Outcome
The plugin gained significant recognition, being selected as one of the top 100 plugins, and has garnered a user base of 20,000 individuals. Currently, the plugin experiences a daily usage of 100-200 users, demonstrating its successful functionality across all operations
Invoice Classifier
Technologies Used
AWS Tech Stack (Textract, Comprehend, SageMaker), GroundTruth), Python
Client Objective
To automate manual process of enriching commercial truck ads from invoices
Outcome
The project achieved the outcome of automating the manual process of enriching commercial truck ads from invoices, resulting in reduced recurring digital human labor. The solution provided a cost-effective and scalable approach, meeting the client's objective.
Local Databricks Deployment
Technologies Used
Databricks Dolly, Huggingface Transformers
Client Objective
At BestViewsReviews, our aim is to assist users in making well-informed purchasing decisions by analyzing product reviews and providing ratings for each analyzed product. Additionally, we tackle the challenging task of extracting key value pairs from raw data to display various product specifications. Initially, we employed GPT3.5 for this purpose, which proved effective but costly. During our exploration, we discovered Dolly, a language model, and conducted experiments by fine-tuning it on our own dataset using the transformers library. We successfully deployed it on our in-house machine equipped with an RTX3090 graphics card.
Outcome
The results were outstanding, yielding a remarkable accuracy of 92% compared to GPT3.5. Consequently, we transitioned from GPT3.5 to Dolly, resulting in significant cost savings by reducing OpenAI API calls.
Valuation Report Generator
Technologies Used
OpenAI, LangChain, PandasAI, Matplotlib Agent
Client Objective
Create a valuation report and summary of key financial metrics.
Outcome
Client has a database of key industry and organization specific financial and valuation metrics. Comparative reports and charts need to be generated from organizational knowledge of these metrics. Guard rails were built to stop the model from hallucinating
FAQ & QnA Generator
Technologies Used
OpenAI Chatgp3, Python
Client Objective
The client has the amazon product reviews and from the reviews, we have to generate the FAQ and QNA. Using the OpenAI gpt3 and prompts created the FAQ and QNA for the client
Outcome
Leveraging OpenAI's GPT-3 and customized prompts, we developed a solution that automatically generated FAQ and Q&A content for the client. The outcome of this project is that the generated FAQ and Q&A content is now live and available in production on the client's platform, specifically in the BVR system. This enables users to easily access relevant information and find answers to their questions based on the product reviews
Digital Catalog building
Technologies Used
Python Scripting, Data Engineering, PHP Development
Client Objective
To build and maintain a database of automobiles which required data gathering from more than 1500 OEM websites.
Outcome
Our Python and manual data-scraping team utilized in-house developed scripts to improve efficiency in collating data. We are now able to process ~3X OEMs at roughly one fourth of the cost
Reducing Turnaround time
Technologies Used
Open source software, Machine Learning, Root Cause Analysis Research
Client Objective
To reduce turnaround time for customer queries.
Outcome
An open source customer ticketing solution was deployed which increased visibility of issues and Machine Learning based solution was deployed which automatically categorized tickets and assigned them to the right customer service agent. The turnaround time has now been reduced by 60%
Specification Extraction
Technologies Used
OpenAI GPT4, Dolly v2, Selenium, Deepspeed, Cuda, LabelStudio
Client Objective
Automate extraction of specification (key value pairs) from unstructured product name and features written by manufacturers.
Outcome
Exploratory data analysis pipeline setup to extract the information using GPT4. Data labelled on label studio and Dollyv2 was finetuned on Hybrid cloud to reduce cost.
Auto Agents for Lawyer
Technologies Used
GPT4 APIs, LangChain, Agents, Prompt Engineering, Semantic Search
Client Objective
Legal judgement prediction for litigations.
Outcome
Implemented LangChain agents that can semantically search for information from a legal document, summarize, and highlight the key terms. Auto agents argue for and against the case. After arguing for a certain limit, the probability of win/ loss was predicted.
Product Detail Page Generation
Technologies Used
OpenAI, Semantic Search, HuggingFace BERT/ BART/ Pegasus
Client Objective
Summarize customer reviews on a specific aspects for a category.
Outcome
Implemented a complete Data Engineering pipeline to extract reviews, semantically search relevant sentences from text, analyse the sentiment, and summarize it using Google Pegasus on Hybrid cloud. Data annotation was assisted with OpenAI.