Purushottam

System Design & Leadership

• ›Architected a multi-tenant CRM end-to-end — led system design, frontend decisions, and a 5-person team across deployments for U.S. fleet dealerships
• ›Designed per-tenant data isolation: each company gets its own customer base, agent roster, contract ledger, and comms history on shared infrastructure with independent scaling

Scalability & Async Infrastructure

• ›Managed all webhook events and background jobs via RabbitMQ message broker with Dramatiq workers — bulk import, campaign scheduling, and workflow automation run fully async
• ›Built a bulk import pipeline (CSV/Excel) with deduplication, field mapping, and auto-triggered onboarding workflows — no manual intervention required at scale
• ›Engineered a Zoho email fetcher that syncs inbound customer emails into the CRM in real time via webhook-driven event processing
• ›Scheduled SMS/email campaigns with custom frequency controls — configurable per tenant, per campaign type, and per customer segment

Communication & AI Integrations

• ›Built a sequential bulk calling system eliminating manual dial friction for dealer agents — staff can run outreach campaigns at scale with one click
• ›Engineered a conversational AI bot enabling 24/7 customer coverage — handles inbound queries, follow-ups, and escalations without human intervention
• ›Delivered end-to-end payment integration with Twilio, SendGrid, and Jupico — automated voice follow-ups, installment tracking, and real-time sync
• ›Built custom Claude plugins and automation integrations directly into the codebase — streamlined internal engineering workflows

Ownership & Impact

• ›Owned production issue resolution across the full stack — from API bugs to infra incidents — with zero-downtime deployments via Docker on Apache/AWS
• ›Unified dealer portal replaced four separate tools; manual overhead cut by −60%, engagement up +30%, operational efficiency up +40% across all tenants

How I Assisted

• ›Supported faculty in delivering Python and DSA courses to undergraduate cohorts — lecture support, lab sessions, and grading
• ›Ran weekly one-on-one mentorship sessions for students who needed extra help with programming fundamentals

How I Helped Students

• ›Answered questions in and outside class, breaking down complex algorithmic concepts from first principles
• ›Extended two years of active competitive programming mentoring at Code@Amrita into a formal TA role

Recognition

• ›Awarded Green++ — the highest student performance grade at Amrita School of Computing
• ›Communication and documentation skills built here carried directly into MaTrack technical writing and ICVR 2025 research presentation

What I Built

• ›Neonatal care web application for Amrita Hospital (Kochi) using Frappe — digitised clinical data capture workflows, replacing manual documentation and reducing data entry load by ~35%
• ›Natural-language-to-SQL query system eliminating traditional filter-based search for clinicians — enabling intuitive data retrieval 3× faster across a highly variable patient dataset

How I Trained the Data

• ›Fine-tuned Llama-2-7B on ~3,000 neonatal attributes — engineered schema-aware prompt pipeline so the model generates valid SQL with the Frappe database as context
• ›Implemented similarity-based patient clustering for early diagnosis — surfaced infants with matching disease traits automatically

Results

• ›Diagnostic identification time cut by ~40%, reducing delays in treatment workflows
• ›Data entry load reduced by ~35%; clinicians retrieve records 3× faster than filter-based search

What I Learned from Scratch

• ›Rapidly acquired Unity, C#, and MRTK from scratch — applied 3D spatial mapping, physics, and real-world alignment within the research window
• ›Integrated photogrammetric reconstructions of Moovar Kovil into HoloLens 2, solving frame rate bottlenecks from ~15 fps to a stable 60 fps through scene optimisation and baked lighting

Research & Design

• ›Researched the architectural and historical context of Moovar Kovil to design interactive MR learning experiences
• ›Iterated across multiple feedback cycles to build KovilLens — a cultural heritage preservation platform for digital exploration of a 9th-century Chola temple

Publication

• ›Contributed to a peer-reviewed paper accepted at ICVR 2025 (Wageningen, Netherlands) — translating hands-on engineering and domain research into a conference-ready paper within a 3-month research window

What I Built

• ›talkToLocals — full-stack cloud-native platform connecting travellers with local guides based on location and interests
• ›Django REST backend on Azure App Services with ReactJS frontend, Azure SQL for data and Blob Storage for media
• ›Azure Functions for serverless event processing — confirmation emails, match notifications, async tasks

What I Learned

• ›End-to-end Azure architecture across five services — App Services, Functions, SQL, Blob Storage, and ML
• ›CI/CD pipelines, environment configuration, monitoring dashboards, and cost optimisation on cloud infrastructure
• ›Integrating Azure ML recommendation models into a production web application

Achieved

• ›Azure ML-powered guide matching boosted communication efficiency by +30%
• ›Delivered a fully deployed, production-grade cloud application within a 4-month apprenticeship
• ›Earned 187 badges and 39 trophies on Microsoft Learn — completing the full Microsoft Azure Apprenticeship track