Summary
Software engineer with applied AI and data work in production: a computer-vision pipeline that processed 90,000+ exam sheets, an incremental DuckDB analytics pipeline (42 minutes down to 3.8), and a retrieval chatbot over PDFs. Currently at Runo, and studying PyTorch and generative AI.
Experience
Software Engineer
On a sales CRM, I take the problems that cut across the product and make them dependable: rep chats that never reached the CRM, analytics nobody could trust, calling modes an account couldn't combine, and a daily report run that took 42 minutes.
- WhatsApp Web CRM extension: Sales reps work their leads in WhatsApp Web, and none of those conversations reached the CRM, because WhatsApp offers no API for it. I built a Manifest V3 Chrome extension that finds WhatsApp Web's internal data stores at runtime and syncs chats, contacts and call history in the background, without ever switching the chat the rep has open.
- Integrations platform: Every CRM connector in the admin app was its own one-off screen, so each new integration meant rebuilding setup, validation and error handling from scratch. I replaced them with one setup flow they all share. Zoho, Salesforce, HubSpot, Facebook Lead Ads, WhatsApp Business, AWS S3 and Google Cloud Storage run on it today, and the next connector starts from it instead of from zero. 7 integrations
- Product analytics: Tracking calls sat directly on the Amplitude SDK inside screens, so staging events leaked into production dashboards and re-renders fired duplicates. I replaced them with one typed analytics layer that keeps test data out, drops duplicates and buffers busy activity on the device. It carries ~15M events a month, and has reprocessed 45M in a single run. ~15M events/month
- Call-analytics pipeline: The daily call-analytics run recomputed every file from scratch and took 42 minutes. I rebuilt it as an incremental DuckDB pipeline that only processes what is new. It now finishes in 3.8. 42 min → 3.8 min
- Hybrid calling: An account could use SIM call tracking or cloud calling, never both, and changing that meant reconfiguring the account. I built the session handling that keeps both live, so an agent switches mode mid-shift without logging out, and replaced the on/off flag with per-call tracking so reports show which path each call used.
Mobile Application Developer
Part-time during the last three semesters of my B.Tech: rebuilt NMOLD's inventory system and connected its factory machines to it.
- Inventory system rebuild: NMOLD's inventory ran on a rigid legacy Python/Laravel stack, and operators typed counts by hand. I led its rebuild in Flutter, Express and MongoDB, and added image recognition that categorizes products from a photo, cutting manual data entry.
- Machine data: The plant's Siemens and Delta PLCs sat on an isolated network, so checking a machine meant walking the floor. I built the bridge that brings their data into the platform through OPC UA and Node-RED, and a secured remote dashboard that shows live machine status and can switch equipment. Live at one manufacturing site. 1 site, live
Campus Software Developer
Paid, part-time during my B.Tech. As the college's only developer I built and ran its exam, feedback, ticketing and bus-tracking systems, and kept them running until I graduated.
- College portals: Built and maintained the feedback, RISE, PIET Quest, KBE quiz, ticketing and payment (HDFC Bank), WhatsApp messaging, and bus tracking and attendance systems.
- Exam integrity: Campus exams ran side by side in shared labs, where copying was easy. I gave PIET Quest a randomized question and option order per student, and the order reshuffles when the portal detects the student leaving the exam window.
Freelance
Automation and AI
A Python OMR reader that processed more than 90,000 exam sheets, plus Jenkins and Terraform work. Supplier price scraping and margin-based price updates for an OpenCart store. A chatbot over product documentation on AWS Lightsail, and a scraper for NGO details.
Blockchain
A blockchain donation tracker for an NGO, an OCR answer-sheet checker, and client websites and hosting.
Products for startups
A job portal and equipment marketplace for the construction industry. A hospital management system, built with a student team. A social platform where students track and share achievements. The frontend website for a milk-delivery startup.
Sites and portals
An exam module in Laravel for a civil-services coaching institute. A quiz portal for intern certification, and updates to the company website. Upkeep of a legacy CodeIgniter property listing site, and a Node.js information portal. Fixes and improvements to a company website.
Mobile apps
A Flutter redesign of a prayer app for Ramadan, with audio playback. An Android news app, plus a rentals platform and a classifieds website.
Hosting and operations
Plesk servers, DNS and security, including honeypots, across a web agency's hosting. Hosting for a news website, including its move to AWS Lightsail. Recovering client sites from backups, then hosting, DNS and email.
Education
B.Tech, Information Technology (AI & Data Science specialization)
Data Science Training Program
Projects
OMR Sheet Automation
Exam answer sheets were checked by hand, and simple scanners broke on skewed or smudged scans. I built a staged Python and OpenCV pipeline that reads them and resumes from failures instead of rerunning. It has processed more than 90,000 sheets and is still in use.
RAG Document Chatbot
Retrieval-augmented chatbot over PDF docs: embed with a hosted model, retrieve top chunks, answer with an LLM constrained to that context. Flask UI, and a clear path from one shared index to per-user indexes.
No-Code ML Environment
Flask app that wraps common supervised-learning pipelines behind a UI: CSV upload, column picker, sklearn regressors or Keras LSTM, then MAE/MSE/RMSE plus a prediction plot.
Product Analytics Pipeline
Tracking calls were scattered through Flutter screens, so staging events polluted production dashboards, re-renders fired duplicates and offline activity was dropped. I built one typed analytics layer over Amplitude that fixes all three. It carries ~15M events a month and has reprocessed 45M in one run.
Hopfield Network Visualization
Two Jupyter notebooks exploring Hopfield networks: a basic 2-neuron energy landscape demo and an advanced 5x5 grid showing pattern storage and noisy-input recovery with animation.
E-commerce Price & Catalog Sync
An electronics store checked supplier prices by hand every day. I automated supplier catalog extraction and price-drift detection, with margin rules and outlier thresholds that produce ready-to-import OpenCart files.
Manufacturing Inventory Platform Rebuild
NMOLD's inventory ran on a rigid legacy stack: new product attributes needed schema migrations, operators typed counts by hand, and machine data never left the factory network. I led the rebuild in Flutter, Express and MongoDB and bridged the plant's PLCs into it. Live at one site.
WhatsApp Web CRM Sidecar
Sales reps' WhatsApp Web conversations never reached the CRM, and WhatsApp has no API for it. I built a Manifest V3 Chrome extension that finds WhatsApp's internal stores at runtime, syncs chats and call history in the background, and never changes the chat a rep is looking at.
Skills
PythonLlamaIndexRAGOpenCVscikit-learnKeraspandasNumPyDuckDBFlaskSQLAWS