Home

Nitin - AI/ML Architect

vinny@uniteditinc.com

Location: San Francisco, California, USA

Relocation: Yes

Visa: H1-B

Nitin Singhal 469-575-4104 vinny@uniteditinc.com https://www.linkedin.com/in/nitinsinghal03/ | https://datawisdomx.com/ | https://github.com/datawisdomx1 I have over 19 years experience as analytics professional. I want to help companies outperform their industry by building breakthrough products using AI ML technologies and data science. I follow arXiv, blog about AI & publish models on GitHub SUMMARY AI ML Engineer: Gen AI, RAG, Prompting, CNN, RNN, Deep Learning, Reinforcement & Transfer Learning, Regression, Classification, Clustering, Recommender systems, Time Series Forecasting, Optimization, Synthetic data, Churn, A/B tests AI ML Architect: Solution design, platform implementation, models, cloud, data pipelines, collaboration, communica-tion Machine Learning: Model tune, scale, deploy, maintain, APIs, flask, MLOps, Databricks, Sage maker, CI/CD, Docker, EDA NLP: LLM, Autoencoder, Transformers, BERT, GPT, Llama, Hugging Face, RNN, LSTM, spaCy, NLTK, Topic, Sentiment, POS Gen AI: LLM, Architecture, tuning, alignment, SFT, RLHF, RAG, Prompting, CoT, ToT, CUDA GPU, Quantization, Langchain, LangGraph. Multimodal: Text, speech, image, numeric model using RAG, llama, whisper, OpenCV, tree model, synthetic data Computer Vision: CNN, ViT, YOLO, GANs, OpenCV, LayoutLM, OCR, Image classification, Object detection, Livestream Programming: Python, Numpy, Pandas, Scikit-learn, TensorFlow, Keras, Pytorch, Matplotlib, Solver, Pulp, Java, Excel Data Engineer: ETL, Pipeline, Big data, SQL, Big Query, PySpark, Snowflake, Vector DB, MySQL, XML, JSON, MongoDB Mathematics: Statistics, Probability, Linear Algebra, Calculus, Optimization, Algorithms, Data structures Business Analysis: Requirements, Specifications, Cross functional, User Stories, Dashboards, Process flow, Tableau Software Development: Coding, Full lifecycle, agile, UI, deployment, mobile apps, Cloud (Azure, GCP, AWS), Git, JIRA Project Management: Leadership, communication, plans, budgets, KPIs, stakeholders, governance, risk management PROFESSIONAL EXPERIENCE AI/ML Architect. Contractor. Jersey City, NJ, US Aug 24 Current Hexaware (AxaXL): Generic question answer model to retrieve all possible answers for a given question from docu-ments Architecture, requirements, data engineering, RAG, LLM, FAISS vector db, retrieval, Databricks, python, PySpark, SQL Model training, LLM embeddings, vector db indexing, similarity search retrieval, db search queries, UI output, testing Applying AI ML, Deep Learning, LLM, NLP algorithms, big data, data engineering, spark clusters, Azure Devops, Git Other data science projects for different clients, requiring end to end solutions from ideation to model to deployment AI/ML Engineer. Grid Dynamics. Greater Boston, MA, US Jul 22 Jun 24 Client (Verizon): Predict customer churn in real time with marketing call centre agents and make retention offers Real time churn prediction & offers using AI Deep Learning NLP on transcript text, model parameter tuning in GCP Custom transformer model processing real-time text to predict customer churn probability using NLP techniques Implemented prompt engineering, memory management, and retrieval-augmented generation (RAG). Designed workflows for multi-agent AI systems using LangGraph. Tokenization, pyspark UDFs, StructType schema operations, tensorflow numpy operations, hyperparameter tuning Rebuilt data pipeline using Google Speech To Text, multi timeframe sampling & batch processing to extract call text Optimized read/write operations for BigQuery Google cloud API OAuth timeout issues due to large data size. Developed and optimized AI applications using Langchain and LangGraph. Multiple data quality reports using BQ queries (nested data, SQL) for token quality, recency and relevancy analysis Multiple client presentations, excel reports to present model results across different parameters and timeframes Client (Verizon): Forecast & Optimize wired networks call volume & engineer scheduling with time & headcount limits Multi constraint linear optimization and stacked ensemble time series model to forecast volumes & schedules in GCP Custom mixed integer programming cost and volume demand optimization model with slack variables for constraints Prophet time series and XGBoost ensemble model call centre volume forecast using historic data for multiple products Google cloud Kedro pipelines, call & engineer schedules demand forecast at hourly weekly level with multiple con-straint YAML files Client (Verizon): Architect model merging 3 products (wired network, construction, maintenance) for workforce skill sharing, schedule call volume & cost forecasting & optimization, with time & headcount limits, data engineering, re-ports Multi constraint linear optimization and AI ensemble time series model to forecast volumes & schedules in GCP Data: Engineer & pipeline demand & forecast data (>10mn rows, >100GB), weather, calls, engineer skills, KPIs, clusters BQ Data engineering for end to end data flow with all input sources, elements, types, intermediate enrichment, prod-uct configurations, wrangling, model and optimization prediction outputs, KPI calculations, report and schedule gener-ation ML Engineering, business logic extraction, legacy model maintenance, analyzing log files, python bug fixes, JIRA, SCRUM AI ML Engineer & Founder. Mobicloudtrees Ltd. London, UK. Delhi, India Feb 12 Jul 21 Founded DataWisdomX, a data science platform providing training courses, expert matching & machine learning mod-els Course with theoretical math s, regression, classification, clustering, deployment, AWS, Sagemaker, End 2 End model Multiple Deep learning models in ANN, RNN, DNN, CNN, FCNN, BERT, LSTM, Recommender systems, Finetune LLM Llama, computer vision, Langchain, CUDA using Keras, Pytorch numpy, python tensorflow, pandas, SQL Time series forecasting (ARIMA, SARIMA), Imbalanced data (SMOTE), Model stacking (Vecstack), Genetic algo-rithms, PCA, LDA, synthetic data generation, optimization solver, EDA, statistics. Code and results in github Founded CurrentiaX, B2B currency platform, algorithmic prediction, low price execution, AML, RegTech, cross border Deep learning regression predicting currency prices using macroeconomic central bank and forecasted data in ma-jor pairs (USD, GBP, EUR, JPY) using python, pandas, numpy, scikit-learn, xgboost, ANN, NLP, matplotlib, JSON, REST API Partnerships with Execution providers, SME platforms & agencies, RegTech firms and media agencies Built a low cost Robo Advisory platform for small investors multi-asset portfolio using machine learning algorithms Regression prediction of asset prices, forward optimized portfolio (stock indices, commodities, currencies, real es-tate, bonds) weights risk return, auto rebalancing using python, pandas, numpy, xgboost, ANN, Sagemaker, joblib Automatically construct, re-balance a multi-asset portfolio using predicted asset prices for small/medium investors Advised KNAB Finance on predicting loan approvals and credit risk, using machine learning models Worked with founders, credit, technology teams to define business problem, process and data exploration Architect design deep learning, regression, clustering models for delinquency buckets, seasonal patterns in GCP EDA statistics, Data wrangling, python, numpy, pandas, SQL, regression, clustering for segments, time series for loan trend/seasonality, management reports, visualization Advised NOMISMA s founders on cloud accounting product, multiple projects with software & product improvements Multiple projects with full lifecycle BRDs, algorithms, test cases, UI design and improve software quality Bookkeeping & final accounts, Corporation Tax & VAT returns, Payroll, Self-Assessment, Making Tax Digital R&D, XBRL, API & data mapping, multiple offshore teams and integration with external vendors software for Pen-sions Auto Enrolment (pensionsync), FX feeds (openexchangerates) and Dividend feeds (google finance) Entoss Technologies Business development & fund raising project for process automation startup Worked with founders & technology teams to build product marketing documents, pitch deck and website Business development, lead generation, sales prospecting, pitching to investors, product reviews, pricing strategy Founded mFoodx, a B2B cloud marketplace for wholesale food industry, connecting buyers, sellers & logistics firms Product - Order management, market demand/supply, outsourced logistics, machine learning models, messaging Business & product development, sales & marketing, trade shows, pitching to investors, lead generation Developed a pioneering mobile & cloud-based analytics application for healthcare technology industry Researched & designed a Health & wellness platform for Dr s and patients to analyze disease & lifestyle data Business Analyst/Project Manager. NS Advisors. London, UK Sep 08 Jan 12 Led analysis & design of Delta-1 systems at Nomura. System migration & transformation of swap ETF & SBL products Transformed Equity Finance & Delta-1 systems at HSBC for equity swaps pricing, funding & automated client reports Traded macro & momentum strategies-based futures in major currency pairs, government bonds & stock indices Sr Business Analyst/Project Manager. JPMorgan Chase. London, UK Aug 05 Aug 08 Led global technology development on equity swaps & funds derivatives products with budget of over $1 MM Global distributed users and requirements, common solutions, workshops, technical product strategy & specification with business heads, trading, IT, operations, finance, sales, risk, quant and IT client on-boarding Equity swaps settlements portal, multi-asset class funds portfolios, systems and process reengineering Consultant. Accenture. London, UK Aug 04 Jul 05 Designed LSE s trading system SETS. Quality manager for design of Lloyds TSB Registrars Corporate Actions product Associate. Barclays Capital. London, UK Aug 03 Jul 04 Developed & designed trading system and bond pricing for Funds derivatives portfolio of CPPI & Leveraged Certificate s Consultant & Software Engineer. OSI Technologies. Hyderabad, India Jul 99 Jun 00 Developed email marketing product, restructured services unit ($200k saving) and strategic analysis for digital marketing KEY AI/ML/Gen AI PROJECTS AI platform for providers & partners to manage healthcare lifecycle & enable future investments Problem: Multiple data sources & manual processes prevent providers & partners from managing patient health lifecycle (disease, care, lifestyle) efficiently, increases cost, reduces providers available capacity and prevents future investments Data: Numeric/Text/Image/Audio - Prescription, EHR, Labs, Dr, Wearables, Insurance, Pharma, Public, Synthetic Solution: HIPAA compliant healthcare lifecycle AI platform (multimodal DNN, CNN, ViT, NLP, Gen AI, Tree based), ana-lytics API, enable investments, Databricks, SQL, Python, metrics Using Databricks with AWS for MLOps process, experiment tracking, logging, versioning best model during training, storing the artifacts in Unity Catalog (dbfs, metastore, schema), cloud buckets API, audit, governance, policy features Process Insurance Claim using GenAI OpenAI RAG Vector db(FAISS) Llama3 Langchain Extract claim amount, relevant data from insurance claim using GenAI LLM model without need for training Using RAG, vector db for document retrieval, enhancing user query context for chained prompting of llama GenAI model No need for fine tuning LLM while using latest information from database/documents for relevant recent data Marketing Models Gen AI ML models for product attribution, customer churn, marketing mix, supply chain optimiza-tion Heuristic models (multi, first, last, decay, touch distribution), topic modelling, sentiment analysis, wordcloud using py-thon, pandas, numpy, nltk, pytorch, keras, tensorflow, scikit-learn, seaborn, matplotlib Gen AI, NLP, LLM, recommendation system models to predict customer demand behavior, segmentation, targeted marketing collaborative filtering, frustration from feedback text, call center transcript, social media views Using Vader sentiment analysis, LDA topic model, KMeans clustering, BERT, Gen AI fine tuned models, multi-layer neu-ral network, RNN LSTM for modelling past sequential data, cosine similarity, one hot encoding, feature engineering, metrics Product attribute extraction using CV (computer vision) CNN Gen AI model Architecture and implementation of pre-trained Faster R-CNN model with a ResNet-50 backbone on COCO dataset to extract product attributes by training custom product image data using object detection, identifying object labels from the bounding boxes, performing Gen AI NLP operations on label product attributes CNN architecture- Input Layer, Convolutional Layers, batch normalization, Residual Blocks, Pooling Layers, Fully Con-nected Layers, Output Layer Using pytorch, torchvision, SGD optimizer, small samples, epochs, reduced training loss, RTX4090 GPU s 48 GB RAM Generating Product Description SEO CTA Twitter Keywords from Images using GenAI API Prompting Product image (retail) passed in OpenAI API prompt for description, used for another API prompt for detailed descrip-tion Multiple different API prompts to extract different types of keywords for SEO, CTA, twitter from the description HTML tags with all information for automated webpage generation OpenAI gpt-4o API calls, GPUs (Q RTX 8000 48GB), pytorch, python. Model can be scaled for millions of product images Albertsons Retail grocery store order fulfilment forecast & optimization Problem: Improve order fulfilment shortage at stores from distribution centers and vendors across all products Data: Numeric, limited (2 years, few vendors/distributor/stores/products), no causal (transport, weather, etc) fea-tures Solution: EDA, regression models to predict future order quantity, optimization models for ideal factors to get 100% fulfillment, Time series analysis for trend/seasonality, dashboards using Excel/Tableau, Synthetic data to improve models Central bank interest rate decision using reports Problem: Predict central bank interest rate decision (hike, no hike, cut) by analyzing monetary reports text data Data: Small, monthly for 16 years, only for US Federal reserve meeting minutes. Manual tagging of interest rate deci-sion Solution: NLP classification and n-gram (different combinations) models using NLTK with 60%-80% accuracy Insurance claims cost Problem: Predict customer insurance claims cost(severity) to company Data: Large with 180k rows, 130 features, numeric and categorical, good quality Solution: Different regression models (ANN, LSTM, XGBoost, RandomForest), compared on accuracy with hyperpa-rameter tuning, more ANN nodes & layers Livestream image object detection & count Problem: Count the number of people in a queue using live camera feed using computer vision model Data: Livestream camera data under different environments (indoor, outdoor, different lighting, movement) Solution: Computer vision model analyzing livestream frames using OpenCV, YOLO5, COCO dataset, and pytorch EDUCATION MS Business Analytics. W. P. Carey, Arizona State University. Tempe, Arizona, US. Aug 21 May 22. CGPA 3.88/4 MBA (Finance, Systems), Indian Institute of Management Calcutta, India, Apr 01 Mar 03 B.E. Computers, Pune University, India, May 95 May 99. Class topper in 2nd and 3rd year. Gold medalist Cover Letter Dear Sir/Madam AI ML technologies are disrupting the way we approach business problems in most industries. The ability to algorithmically analyze large amounts of disparate multimodal data and make useful suggestions in real time is allowing people and com-panies to generate better and more creative solutions for a variety of problems at a lower cost. The rate at which new im-proved algorithms and technologies are being created in AI is further accelerating this trend. Your company s objectives are future growth oriented in applying AI ML analytics, a field I d like to work in. My deep commercial experience in the latest AI/ML technologies across multiple industries and countries enables me to provide creative & practical solutions for a variety of problems. All the startups I founded were solving large real world problems impacting businesses and people in a positive way. They were well ahead of the market and replicated later by some competitors. Regards Nitin Singhal San Francisco, CA Keywords: continuous integration continuous deployment artificial intelligence machine learning user interface materials management database rlang information technology microsoft California Massachusetts New Jersey