 Led the design and implementation of agentic systems and voice user interfaces (VUI) using Next.js (React based) frontends/BFFs and FastAPI backends for completions, significantly enhancing operational efficiency for work-flows.that were previously resistant to digitization and automation. Mentored engineers and did work on handling streaming data, providing example code and guidance for React-based.frontends, enabling successful integration of real-time features and improving team capabilities. Fine-tuned a Whisper medium models on synthetic data comprised of industry-specific utterances, increasing accuracy.in recognizing certain industry jargon from a baseline of ~40% to over ~80%. Deployed systems and setup CI/CD pipelines using Github Actions, Docker and AWS ECS. Designed and engineered real-time, bi-directional streaming data pipelines using WebSockets to process text/audio input, perform transcription via Whisper, leveraging OpenAI API and LLMs finetuned using DPO for reasoning, and stream responses back to clients, ensuring a highly responsive user experience. Leveraged advanced prompt engineering techniques and implemented function calling (tool use) to significantly reduce LLM hallucinations, resulting in more accurate and reliable data outputs for critical information.

· Led and implemented a fully functional prototype to improve data security and identity verification processes, adhering to the nascent W3C VerifiableCredential API specification in stable Rust using Actix Web and multiple crypto libraries· Designed and built a bespoke blockchain for secure and transparent supply chain management, ensuring traceability and product authenticity, using Rust and the Substrate framework by Parity Technologies (Polkadot).· Secured a partnership worth over $2 million in potential revenue and 100,000 monthly users, boosting market presence, by developing a fully functional prototype from scratch for image recognition of fashion items to replace manual input.· Constructed an interactive frontend for the prototype accepting an image of the user’s luxury item as input and using a ViT based model to detect brand, color and type with PyTorch, based on the paper "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale" by Dosovitskiy, A., et al. (2020).· Achieved a 85% accuracy rate, along with seamless integration into the website backend with minimal friction and delay for image recognition, resulting in successful deployment into Global Blue, Farfetch and Reflaunt’s systems.· Optimized engineering productivity and software development cycles by mentoring junior and senior engineers, and coordinated API contracts/integration between pricing and other core services in collaboration with the analytics team.· Provided strategic advice to the executive team via architecture planning meetings for platform development, and addressed security-based considerations such as secure authentication protocols and regular vulnerability assessments.

· Increased test code coverage from 40% to 80%, enhancing bug identification and resolution processes, by implementing a testing suite for core entities (model classes) utilizing the pytest framework.· Achieved a 90% reduction in response time and reduced the number of database queries by 65%, enhancing analytics capabilities, by developing a FastAPI-based API for data-related services leveraging asynchronous operations.· Optimized deployment time by 50%, enhancing horizontal scalability, by Dockerizing legacy APIs and implementing docker-compose based orchestration.· Designed and implemented a generic, dynamic way of retrieving, filtering, and returning fields, providing a new migration pathway to sunset legacy APIs using model adapters and leveraging pre-existing data stores.· Enabled the processing of 10,000+ invoices and orders per hour by implementing a distributed task queue to automate generation of invoice/order summaries from externally ingested data, using Celery + Celery beat.· Reduced data ingestion time by 33% for analytics and optimized data ingestion for real-time and batch processes, by implementing a distributed modular data ingestion system for product catalogs using Dask (pandas superset) + Celery.

· Decreased average API response time from 500ms to 400ms (20% performance optimization) and reduced costs, by architecting and implementing a set of entities (classes/models) used in an Async REST API based on FastAPI.· Achieved a 50% reduction in message delivery latency, enabling near-instantaneous user interactions, by developing a real-time communication (RTC) microservice using Websockets + Pub/Sub capabilities.· Reduced the average number of critical bugs per release by implementing a CI/CD pipeline using Gitlab + Pytest.· Minimized the loss of work due to errors, reducing user frustration and wasted effort, by implementing a “save andcontinue” feature in Alfred (an estate planning app) during the creation flow, using React and Django.· Achieved a 10% increase in user activation rate by implementing a brand new on-boarding experience in 2 weeks, using React for frontend development.· Reduced average application crash frequency from 10 a month to 1 instance per month, enhancing system reliability, by developing a full testing suite using Pytest/Pytest-Django, Selenium, and Model Bakery for a web application.· Mentored a backend engineering intern for 3 months to increase proficiency in Python, Django, DRF and PostgreSQL.· Conducted deep analysis of client requirements, ensuring optimal balance between development speed, cost effectiveness, and scalability, leveraging stakeholder interviews and business metrics to inform technical decisions.

· Aided the design and implementation of React components to enhance user search experience and optimize search results, utilizing React Hooks and Material-UI.· Designed and implemented various endpoints to facilitate secure voting processes using Nest.js and MongoDB.· Migrated a highly-customized backend for a combined voting/social app from Nest.js/TypeORM to Django and (DRF), and enhanced system security and user convenience by implementing JWT and social login features.

· Achieved a 13% reduction in customer acquisition and marketing costs by consolidating/cleaning over 2 million records weekly of lead data purchased from a broker, using TowerData API and Python/Pandas.· Boosted marketing ROI by 10% by leading an offshore team of web developers to publish a network of 14+ landing pages based on buyer personas/industries (1,000 daily unique views per page), using WordPress and HTML/CSS.· Increased client retention by 8%, by creating an opt-in SMS marketing program allowing representatives to maintain near-constant contact with funding clients using Python and Twilio API.· Realized a 5% savings in core operational costs and decreased time spent on client-representative communication significantly, by digitizing all forms via the installation/maintenance of the open-source Open eSignForms application.

• Saved 10+ hours of sales data entry weekly by automating the entry of funding client info and documents into Capsule CRM using its API, OmniPage SDK wrapper (for zonal OCR), requests library, and Python. • Improved representative-client satisfaction survey results by 13.29% and significantly reduced hold times, via the design, installation/setup, and integration of a virtual call center into Capsule CRM using Five9. • Saved $50,000+ on commercial real estate rental by deploying various automation and call center systems, enabling remote work and migration of infrastructure to off-premises Active Directory using AWS EC2.