Education
Government Engineering College (GTU), Bhavnagar, India - Bachelor of Engineering (B.E.) in Computer Engineering (2013 to 2017). CGPA: 7.87/10
Technical Skills
Languages & Databases
JavaScript, TypeScript, Python, Go, PHP, HTML, CSS, DynamoDB, MongoDB, PostgreSQL, MySQL, Redis
Frameworks & Tools
Langchain, Langgraph, CrewAI, Node.js, Nest.js, React.js, Next.js, CodeIgniter, Laravel, Vite.js, GraphQL, MinIO, Docker, CI/CD, AWS/Azure DevOps
Experience
Senior Software Engineer (October 2021 to Present)
- Engineered an intelligent document comparison agent for Tata Power's Contract Management System using CrewAI (python) and vector embeddings, reducing contract reviews that previously took 2 days to 2+ weeks per document down to hours.
- Architected an intelligent document summary extraction system for Tata Power's Finance System using LangChain, LangGraph (python), and vector embeddings, streamlining weeks of manual analysis into hours.
- Natural Language to SQL generator using CrewAI (python), SQL Schema ingestion for Unilever's NMCI analysis, achieving 95% accuracy and 50% faster workflow
- Developed an internal document intelligence tool for Codemonk using Langchain.js (Node.js), Vector Embedding, Ollama & MinIO (self hosted file server) that answers queries based on attached documents.
- Developed Nest.js (Node.js) based API modules for ReNew Energy Global following OWASP secure coding practices that digitalized internal workflow
- Architecture shift from Monolith to Microservices to sustain 10k concurrent users with event queue mechanism on Arré Voice app with tech stack Node.js, GraphQL, GO & DynamoDB
- Resolved critical bugs in the RhythmUI (React.js, kitchensink, storybook) component library — fixing the file uploader, build pipeline errors, and Formik integration in the inline editor — improving component reliability and developer productivity
- Optimized QA processes with impactful in-app widget script - BugSnap, that directly adds bugs to Jira with current app state, screenshot or screen recording using Nest.js (Node.js) backend, React.js (Vite.js with RhythmUI) frontend.
- Performance improvement across backend and frontend projects with Node.js, Python, GO as primary tech stack
Web Developer (January 2019 to October 2021)
- Designed and delivered hotel management APIs (Express.js/Node.js) for Shashi Hotels integrating Infor/Agilisys reservations, IoT (thermostat, smart lights, smart locks) controls, in-room services and payments (Stripe, Venmo, FreedomPay) using MERN and Firebase Firestore
- Developed Cookt recipe platform with 10,000+ recipes using CodeIgniter (PHP) and Python scraping
- Engineered Security Bond portal and optimized email templates for Jhaveri Securities using MERN stack
- Integrated Paytm and custom wallet APIs for Vendstop vending machines using Node.js
- Delivered Tugo: meal prep simplified (Express.js/Node.js), Nutrition Calculator (React SPA), and Implicit.io using React, Vue.js, and Node.js
PHP Developer (October 2018 to January 2019)
- Developed web applications using CodeIgniter and Laravel; focused on feature development and UX
- Improved performance and scalability, reducing page load times by 25%
Software Programmer (July 2017 to October 2018)
- Delivered a full ERP for PCB manufacturing covering CRM, BOM, Sales, Inventory, and Shipping using Phalcon (PHP) and MySQL
- Engineered mCare SPA with Node.js, AngularJS stack; ported to cross-platform Electron.js app
Projects
[My GitHub provides a better view of what I've worked on recently.]
Tata Power - Potential Questions
Tata Power finance team needed an automated system were they provides the AGM meeting docs and needs to identify potential questions that can be asked by the meeting attendee. Built a LLM based process, where the documents for the AGM meeting get ingested, parsed to markdown, stored to vector database with it's embeddings and then using series of LLM agents - first an agent process each chunk of the document to identify the potential AGM and stakeholder level questions with few shot prompts, stores the questions in temporary store, perform a deduplication agent on the questions to remove the duplicate or identical questions or merge the questions, and respond with a list of questions with their answer from documents. Parsing done using Docling and OCR models, embeddings generated using amazon titan model, llm agents orchestration done using CrewAI and Python with claude sonnet 4 as the brain.
Stack: Langchain, Langgraph, Agentic RAG, Postgres (pgvector), Docling.
Tata Power - Bank Guarantee Comparison
Tata Power legal team were comparing the multiple versions of bank guarantee with each other or a template manually, going clause by clause. Built a LLM based process, where the documents for the contract get ingested, parsed to markdown, stored to vector database with it's embeddings and then using series of LLM agents - each clause gets extract from document 1, the relevant clause in document 2 gets extracted using that, both gets compared and the observation were sent back along with some other key details extractions. This process has reduced so much back-and-forth in internal process and digitised it. Parsing done using Docling and OCR models, embeddings generated using amazon titan model, llm agents orchestration done using CrewAI and Python with claude sonnet 4 as the brain.
Stack: Langchain, Langgraph, Agentic RAG, Postgres (pgvector), Docling.
Tata Power Contract Management
Tata Power internal team was handling the contracts workflow manually by reading each contract (of 1000s of pages), evaluating and extracting the key details. Built a LLM based process, where the documents for the contract get ingested, parsed to markdown, stored to vector database with it's embeddings and then using series of LLM agents - extract 100s of key details with specified order of precedence in documents. Reducing the manual workload from weeks to hours. Parsing done using Docling and OCR models, embeddings generated using amazon titan model, llm agents orchestration done using CrewAI and Python with claude sonnet 4 as the brain.
Stack: CrewAI, Postgres (pgvector), Docling.
Unilever NMCI Analysis
Unilever NMCI team were looking to automate their calculation and predictions for the cost of their Raw Materials. Their manual process was to review each quarterly or yearly or monthly data in form of excel sheets, identify a pattern and predict the Net Material Cost Impact for each of their items along with some general queries of stakeholders. To solve this, I have migrated the excel sheets to PostgreSQL database, Provided the schema definitions as the context to the LLM (Azure Foundry - Open AI), Built a simple chatbot UI using gradio, and orchestrated series of agent based on the intent of the user message and history to generate an SQL query, run the query and return the result in summarized format.
Stack: CrewAI, Postgres (SQL + pgvector).
Arré Voice
Arré Voice is a voice note first, all inclusive social platform. I had appointed as the Senior Backend Engineer at Arré Voice, My earlier tasks were to complete the backend features and deploy it, which included managing ArangoDB schema, various modules' APIs like Post management, User management and Feed algorithms based on the users connections and preferences. Later due to Latency issues, we planned to migrate the existing monolithic backend system into a module based micro services having separate services for Feed, Post management, Creator Studio, User management, Notifications, Logging etc. all these services could talk to each other as and when needed using asynchronous Published and Subscriber mechanism.
Stack: Node.js, Apollo GraphQL, DynamoDB, NeptuneDB, Go, Redis, Google Pub/Sub, AWS ECS, AWS CDK.
ReNew Energy Global - Tender Management
Built the backend APIs for Renew Bidding Portal internal team to move the offline and tedious tender management process to online. Used Nest.js as backend API server. Deployed on Azure VM with private VPN on Intranet with docker. Implemented the security practices as per OWASP guidelines to make sure the system is secure from all the vulnerabilities.
Stack: Nest.js, MongoDB, Azure Entra, OWASP security practices.
Shashi Hotels app
Shashi Hotel app - is the all in one app from reservations to in-room services, from in-room device control to loyalty program. Built the APIs for the mobile application and Admin panel using Express.js and Node.js with MongoDB as database. Also, built the Admin Panel frontend using React.js. Integrated multiple 3rd party APIs and SDKs like InforHMS & Agilysys (reservations), Assaabloy (Room Key Generation), Stripe, Venom and FreedomPay (Payment Gateways) along with handling concurrent reservations and slot based meeting room reservation handling.
Stack: Node.js, Express.js MongoDB, distributed services.
Vendstop
Developed APIs for Vendstop smart vending machine application, enabling automated door access via dynamic QR codes and implementing Paytm wallet integration with minimum balance requirements. Engineered backend services to validate wallet sufficiency prior to door unlock events and integrated with the machine's 360-degree camera system and pre-configured item pricing database for inventory tracking and transaction management. Door events and product detection were managed by the vending machine manufacturing team; I created wrapper APIs to facilitate seamless integration.
Stack: Loopback.js, Node.js, MySQL, Paytm.
RockED
RockED is a micro-learning platform designed for car dealership employees to enhance their product knowledge and sales skills. I developed analytics APIs that track user interactions, learning progress, and content engagement. These APIs provide insights into user behavior, allowing the platform to deliver personalized learning experiences and identify areas for improvement in the training content. The analytics system was built using Node.js for the backend and PostgreSQL for data storage, ensuring efficient handling of large volumes of user data while maintaining scalability and performance.
Stack: Node.js, PostgreSQL.
Miscellaneous
- Multiple personal and professional web apps and backends: Cookt (recipe platform, CodeIgniter/PHP, Python), Jhaveri Investify (Laravel), Jhaveri Security Bonds (MERN stack), Tugo: Meal Prep Simplified (Express.js, Node.js, MongoDB), Nutrition Calculator (React SPA), Implicit (Node.js, Express.js, Vue.js), Fireworks (Node.js, Nest.js, MongoDB), Athena (Node.js, GraphQL, MongoDB), RhythmUI (React UI component library), ERP for PCB Manufacturing (Phalcon PHP, MSSQL) — covering basic to advanced CRUD, integrations, and more.
- Generative AI based side projects: CLI Chatbot & PDF to MD
- Vrindavan Furniture- Fullstack app developed using Next.js, Node.js, Mongodb, TailwindCSS