Where I’ve worked.

• Deployed an LLM-based multilingual transliteration pipeline for 24,000+ named entities, saving an estimated 284 person-days of manual effort within 1.5 weeks of deployment.
• Built and owned a unified Market Processor API handling document conversion and concatenation across 6+ global markets and 5+ formats, achieving a 99%+ success rate across 14,000+ files and 5,200+ document sets.
• Engineered financial document automation pipelines for 14+ stock exchanges, including sourcing agents and a parallelised Excel-to-PDF converter, delivering approximately 10,000 documents to product in a quarter.

As a founding Product Engineer at Falco, I was involved the development of their core products. Our team built MVPs for our products and pitched it to various Multinational Companies and Top Tier Institutes across India, some of them even showing interest to adopt our products.

CaseCrumbs™ - A real-time Decision Lab Platform

• I built and architected the core CaseCrumbs™ system, a Real-Time, Interactive Case Study platform, where a learner’s choices are quantified and graded based on the “rightness” of their choice. This is a decision lab for future-ready minds.
• I built an interactive workflow-based editor and a runtime player for CaseCrumbs™, this empowers even non-technical folks (Subject Matter Experts, Teachers, L&D leaders) to design and make high-end, interactive case studies across various domains, like Recruiting, Business, Technology, Science, Arts, and corporate training modules with close to zero programming knowledge.
• Set up session logging for the same, which tracks the routes taken by the learners, this can be modeled to understand and track learner behavior, both at batch and individual levels.
• I also built a multi-step, multi-stage AI agent using Langchain to develop and generate these case studies. We have successfully generated 500+ case studies in a matter of weeks with over 85% success rate (post Quality Checks). This fully automated AI agent consisted of a retrying system, validity and sanity checks built-into the system.

SustAInSkills™ - Where Every Question is New, So Learning Never Gets Old.

• Architected and deployed a full-stack parameterized learning platform (Next.js, FastAPI, AWS S3) that generates infinite unique question variants through dynamic variable templating, eliminating answer memorization.
• Built an authoring platform (Next.js + FastAPI) with real-time WYSIWYG preview engine, supporting complex mathematical expressions, dependent variable calculations, and multi-format assessments (drag-and-drop, formula validation)
• Keywords: Product Engineering, LangChain, Next.js, FastAPI, React.js

• Engineered a full-stack, AI-powered assessment platform (Next.js, FastAPI, PostgreSQL) to automate subjective grading, reducing evaluation time from days to seconds for large student cohorts.
• Architected a GenAI data pipeline using Python, Google Gemini, and AWS Bedrock to process and translate over 300 unstructured course PDFs, achieving >97% semantic accuracy and preserving complex content like LaTeX formulas.
• Developed a content authoring tool that streamlined a 5-hour workflow into a 20-minute self-service task, cutting human effort by 93% and enabling 40+ SMEs to serve over 2,000 students.
• Designed a flexible JSON meta-framework to serve as a standardized configuration, enabling the authoring tool to support diverse academic modules and question types without code changes.
• Pitched and demoed prototypes to senior leadership, including the VP of Learning Innovation, effectively communicating technical solutions and project outcomes to secure company-wide adoption.

• Collaborated with the Space Astronomy Group to analyze astronomical data and manage large space datasets, order of Petabytes.

• FITS Data Lookup and Extraction - Wrote custom Data Engineering pipelines to extract data from specific X Ray Sources from Astrosat’s database, which may be filtered by criterions such as source and timeframe.

  • Developed pipelines for analysing, visualizing and generating said files.
  • This data may be further used by scientists at ISRO for research and discoveries.

• Transient Detection - Developed logic to detect “X-Ray Anomalies”, which are high intensity sources, for ISRO’s SSM data. Explored both Deep Learning (Auto Encoders) and Statistical Methods (Z-Score) for detection of said transients.

• Markdown based CMS - A markdown based custom CMS for displaying various information related to the SSM Mission.

• Flask Based Web Application - Developed a web portal that streamlined access to AstroSat’s SSM sensor array data, enabling convenient data retrieval and analysis.

• Conducted lab sessions for 30+ students to teach them the working of Linux Operating System.
• Assisted students in understanding the basic commands and functionalities of the Linux Shell.
• Advanced concepts like Bash Programming, file handling and Manipulation, and basic data wrangling using AWK were also covered.
• Evaluated students based on their performance in the lab sessions.
• Keywords: Linux, Shell Scripting, Bash, AWK

• Built a multi-model pipeline for Deep Learning based Biometric Identification of Cattle.
• Extensively Researched both on-field and off-field methodologies to achieve 90+% accuracy on the models developed.
• Helped deploy models on an Azure VM by setting up the required dependencies and environment.
• Keywords: Python, OpenCV, YOLOv8, Azure

• Covered theoretical basics of various fundamental Machine Learning concepts (Linear Models, Gradient Descent, it’s optimisation).
• We applied the learnt knowledge in AI and IOT to build a small project on an object avoidance robot