Ayush K.

Large financial-services group (confidential): Copilot-style AI assistant 1. Designed and built an internal AI assistant used across four channels: web app, Microsoft Teams, Outlook add-in and a browser extension. Now serving ~50,000 users. 2. Built the Python services behind it: RAG pipeline (document ingestion, chunking, embeddings, vector-store retrieval), multi-agent orchestration for multi-step tasks, and tool calling into internal systems. 3. Set up a multi-model LLM gateway (OpenAI and Anthropic models on Azure) with routing, fallbacks and per-team quotas, so models can be swapped without touching the product code. 4. Wrote the Outlook add-in (Office.js / TypeScript) and the browser extension; Angular front end, Node.js API layer. 5. Added an evaluation harness and tracing before scaling up: response-quality evals on real user prompts, latency and token monitoring, and cost controls that cut LLM spend by ~30%. 6. Handled the production side in a regulated environment: access control, data-handling rules, logging and incident follow-up. Stack: Python, Node.js, Angular, TypeScript (Office.js), Azure, OpenAI / Anthropic APIs, RAG, vector store, multi-agent orchestration. Axione (telecom / fiber infrastructure): 1. Led the migration of a core internal service from Django to FastAPI: architecture, implementation plan, code reviews and cut-over. Team of 3 developers and 1 Product Owner. 2. Async FastAPI + Pydantic rewrite brought latency down ~40% and made the service far easier to maintain and test. 3. Rebuilt the business workflows: CSV files dropped by operations trigger multi-step processes orchestrated in Camunda (BPMN), with a React UI for the internal users who follow them. 4. Added observability from scratch: SNMP data exposure from network equipment and Prometheus exporters for the backend services, with dashboards and alerting. 5. Took over an undocumented legacy Django service used by an external front end, as sole developer: stabilized it, fixed the blocking bugs and kept business-critical features running. 6. Built an internal order-management engine end to end (FastAPI backend, React front end), including new backend features and a UI overhaul. Stack: Python, FastAPI, Django, React, PostgreSQL, Camunda, Prometheus, Docker, GitLab CI. Medical Facilities of America (US healthcare): FullStack engineer 1. Built a centralized web application to manage users across several internal platforms and to analyze patient and operations data for the business teams. 2. Designed a centralized authorization layer that puts 5+ third-party systems behind a single secure entry point via Azure API Management, with fine-grained roles and permissions. 3. Migrated the legacy Django codebase to FastAPI, shipping with full end-to-end test coverage so the client could keep releasing safely afterwards. 4. Worked in a team of 3, fully autonomous, with short delivery cycles to meet urgent client deadlines. Stack: Python, Django, FastAPI, PostgreSQL, Azure API Management, React, Docker, GitHub Actions.

Multiple subscription and payment systems

Payment Intent and Checkout Session (SCA, EU Regulations) by tiwariayush

Development of module to stabilize wrinkling and unwanted deflections on inflatable antenna. Inflatable Kapton Membranes are to be used in space structures and their antennas to reduces weight and volume during launch. Development of python module to stabilize inflatable structures made of kapton to be used in satellites and other space equipment. This would help in reducing the cost and cut off extra space being used by satellites and their antennas.

Development and maintenance of intranet applications and Placement portal at IIT Roorkee This group also manages the website and registration portals.

Jupyter Notebooks ares server based notebook used for creating and sharing live code, data reports and visualizations and equations using python kernel as backend. My project was to use ERP5 as the backend and kernel for executing code. This ensures that large out-of-core data stored in ERP5 be readily available and used by Jupyter Frontend.

Reviewed book on Python Data Scraping for PACKT Publishers. The book is based on new scraping techniques using python libraries.