Hey, There
I am Uma Maheshwar Amanchi
Software Engineer
Driven by a passion for technology and innovation, I specialize in creating elegant solutions to complex challenges. With a robust foundation in software development and a keen eye for detail, consistently delivering projects that exceed expectations.
3 Years of
Experience
10+ Projects
Completed
400+
Coding Problems
About Me
Professional Software Engineer(AI/ML) having 3 years of experience
Full-stack developer with experience in building scalable, secure, and user-centric applications on diverse cloud platforms.
Proficient in Java, Spring, data management, mobile and web development, machine learning, and UI/UX design. Leveraged DevOps practices for streamlined development and delivery services
Excels in object-oriented design, agile methodologies, and effective problem-solving. Demonstrates strong expertise in database management and API integrations, with a commitment to crafting maintainable, high-quality software solutions.
Known for exceptional ability to convey complex technical concepts to diverse stakeholders.
Here are a few technologies I've been working with recently:
- Python
- Java
- PyTorch
- Artificial Intelligence
- Machine Learning
- Large Language Models
Where I've Worked
- ArchKey Solutions
- Gainwell Technologies
- Eitacies Inc
- College of Public Health Kent State University
- Accenture Solutions Pvt Ltd
-
Software Engineer
2024 June - Present
Designed and deployed an agent-based invoice processing system using GPT-4o (Azure OpenAI), automating extraction and classification of unstructured invoice data, and accelerating procurement decision-making
Integrated Azure SQL for structured storage and Azure AI Search for semantic retrieval of part details to build an end-to-end MLOps pipeline reducing invoice processing costs by over $1.5M annually and enabling real-time semantic lookup for frequently purchased parts
Built a production-grade conversational AI system using GPT-4o over enterprise finance databases, with Redis caching, Cosmos DB for persistence, and Entra-based OAuth/OIDC authentication to secure user access
Implemented MLOps lifecycle using Azure ML, DevOps, and AI Hub with support for CI/CD pipelines, enabling scalable deployment, experimentation tracking, and model versioning across projects
Mitigated hallucinations in LLM responses by integrating Bing Search agents for grounding and developed a materials classifier using Azure Language Studio to enhance extraction accuracy
-
Software Developer
2023 Nov - 2024 June
Finetuned a Multilingual DeBERTa model to auto-moderate user-generated content in healthcare support portals, ensuring compliance with community standards and reducing manual review time
Replaced legacy BERT model with optimized variant, boosting content moderation automation rate by 15.81% and auto-approving 500k entries resulting in $100,000 savings in data labelling efforts
Curated the training dataset for content moderation using the GPT-4 and increased the training dataset quality and diversity by leveraging the self-instruct methodology
Trained a Mixtral-8x7B Mixture of Experts(MoE) model using Low-Rank Adaptation (LoRA) for content moderation and deployed a 4-bit Quantized Model into production
-
Software Developer
2022 Nov - 2023 Nov
Developed the Natural Language Understanding (NLU) module financial services chatbot by fine-tuning a BERT Transformer model on domain-specific data, enhancing intent classification accuracy by 10% over the previous BI-LSTM approach
Extended English-based chatbot to Multilingual settings by collecting in-domain data for French, German, and Spanish languages and finetuned the multilingual BERT checkpoint using PyTorch framework, Hugging Face library
Implemented a T5 Transformer model to summarize lengthy customer service tickets, achieving a BLEU score of 0.7, thereby improving response efficiency for support teams
-
Front End Developer
2021 Aug - 2022 Aug
Designed an internal plagiarism detection system using Retrieval Augmented Generation(RAG) to detect similar copies of assignments from previous years
Developed system primarily uses the Dense Passage Retrieval (DPR) as the retriever and BART as a generator
Enhanced the overall system by adding a DeBERTa-based Re-Ranker and improved the overall recall@1 by 10 points
-
Application Development Associate
2021 Jan - 2021 Jul
Created an XG-Boost classifier to predict the category of incoming tickets and obtained a strong Recall of 0.95
Designed a Linear Regression model to estimate the total time required to close a ticket and achieved R2 score of 0.87
Leveraged K-Means Clustering and its cluster concepts to identify the incoming pattern of unknown tickets and alerted the on-call team to quickly resolve the rapidly increasing clusters
Portfolio
Some of My Latest Projects