Amit Timalsina Email & Phone Number

Bengaluru, Ka, In

Austin, Texas, Us

Putting a personal financial advisor into everyone's pocket. First hire. End-to-End (AI, Backend, Frontend) productionalization of most important features.• Architect, Engineer, and Product Owner of Chat. Built multimodal chat agent to help with wealth managment, investment queries, and tax planning. Created cost efficient evaluation suite (rulebased, llm-as-judge). End-to-End tracing in Langsmith. Guardrails to protect against prompt injection attacks and system prompt leakage.• Architect, Engineer, and Product Owner of Notification Engine. Hand-crafted 100 different notification templates. Built an event-based architecture to send notifications to users based on their financial data.• Built a chat-based user onboarding experience. The onboarding ended with user getting an answer to a question on their finances. This provided a WOW moment early on (in most cases).• Work on multiple other small features and bugfixes across AI, backend, and frontend.

Mountain View, Wy, Us

SaaS for Impact Creation • Re-FI

Mountain View, Wy, Us

New York, Us

Leading a team of 2 machine learning engineers to build and maintain internal engineering tools for model training, inference, and data assets.▶️ Developed end-to-end pipeline for 100% reproducibility of models and model evaluation metrics by utilizing versioned data store (DVC, GCS, GitLab) and improved logging (ML ow).▶️ Conducted rigorous benchmarking experiments encompassing various document types to validate efficacy of same and cross-doctype transfer learning. When dealing with a limited number of documents, achieved 6% average performance improvement.▶️ Built dashboard to view metadata of data assets scattered through multiple environments. This dashboard helps client api, research, product management, and CXOs as data is our moat.▶️ Was building docsumo evals package for evaluation of kv models, table models, and rule-based.

New York, Us

▶️ Collaborated on building a GPT-based KV extractor to solve the cold start problem for new document types and reducing human annotation by 70%. Implemented zero-shot prompting, chunking, and rule-based strategies to handle hallucinations and edge cases.▶️ Led research efforts for multilingual KV extraction from Bank statements and Invoice, works include ne-tuning SOTA models, preparing benchmarking data, benchmarking models, and creating an inference API for the best model. Resulted in average accuracy improvement of 7%.▶️ Helped in building stronger product by working on new features, feature improvement, and bug fixes of 4 key-value and table extraction apis and 3 packages.▶️ Benchmarked the rule-based header classifier in rent roll table extraction, and added a fasttext based classifier with a 0.96 F1 score to the pipeline.▶️ Performed extensive research, fine-tuned Visual Document Understanding (VDU) models for NER tasks on Invoice, Bank Statements, and Fix forms. Spearheaded benchmarking efforts, which resulted in a remarkable 91%+ accuracy across document types achieved by the team.▶️ Researched and built classification models for the chart of accounts (COA) classification in SaaS, CRE, and balance sheet documents, which currently serve customers in production with an average F1 score of 0.8.▶️ Wrote 7 articles on topics of like OCR, Automated Data extraction, Few-shot learning, and Table extraction that are published in Docsumo's blog.

Plymouth, England, Gb

✅ Worked on DevOps (CI,CD) and Platform automation (CT) for MLOPs using AWS Sagemaker in conjunction with MLFlow.

Kathmandu, Bagmati, Np

✅ As the first AI developer of the company, came up with different ideas to use AI in the company.✅ Built the Delivery failure prediction system for the company which is expected to decrease the delivery failure rate. Implemented CI/CD, Data automation, Platform automation for Automated model retraining.✅ Researched and worked on information extraction (IE) from forms and receipts using OpenCV, pyTesseract, paddleOCR, spaCy, and Transformers.✅ Built the Recommendation engine for eCommerce using Tensorflow and Keras✅ Pretrained a transformer (Bert architecture) on medical data to built a sequence generation model.

New York, Us

Orders completed: 3Profile: https://www.fiverr.com/amittimalsina14Category: DataAchievements/Tasks- Completed three data science projects with a 5-star rating- Worked with NumPy, pandas, matplotlib, seaborn, scikit-learn, TensorFlow, and Keras to do data science projects- Built an end to end machine learning productWhat do I do?- I develop data science and machine learning projects.- I teach Python, NumPy, pandas, matplotlib, and Data analysis.- I do data analysis, data mining, and visualization projects

Palo Alto, California, Us

Conducted webinars on AI and Data Science for high school students in Nepal by inviting professionals in the field.

Lalitpur, Bagmati, Np

• Involved in research, design, review, and refinement of content - reading material, quizzes, assignments, and projects for AWS Data Analytics in Advanced Digital Skills and Global Certification Program under KPYEP.• Taught Data Analytics, Machine Learning algorithms, and amazon web services (AWS) for Machine Learning to different sets of live audiences from Pakistan.

- Being a founding software developer, managed, and documented 3 projects in just 2 months- Worked with a fellow coworker to deliver full-stack multi-platform app in just a month- Worked with React-native, Django, MYSQL, technologies to built full-stack applications

Lalitpur, Bagmati, Np

AWS AI/ML Interestship program. The AWS AI/ML Interestship is a 4-month interestship program. This is a guided learning program and Mr. Anjani Phuyal will be guiding students during this program. Mr. Anjani Phuyal is an AWS APN Ambassador and AWS Certified Machine Learning.The target group for this Interestship is Final year students and graduates. However, you have to dedicate 25+ hours per week during this program.

Achievements/Tasks:- Worked as Teaching Assistant for Machine Learning- Conceptualized, planned, launched, and led the Young Scientists Community for high school students.- Worked with other researchers and did extensive data analysis on existing datasets, also learned about research methodology.

It was my first job experience during my summer vacation of grade 11(Junior Highschool).-I designed 2-3 brochures.-I designed 2-3 logos.-Some other designing stuff.

Engineer

High School Diploma, General Science

Middle School Diploma, Computer Engineering