<aside> <img src="/icons/bullseye_gray.svg" alt="/icons/bullseye_gray.svg" width="40px" />
Hey 👋, I am Pranshu. I am a Full Stack Developer at Cosight.
I like understanding how systems work under the hood; databases, networks, queues, performance and then using that knowledge to build straightforward, practical solutions.
</aside>
Full Stack Developer — Cosight
Oct 2025 – Present
Engineered a robust multi-source ingestion pipeline to aggregate and normalize disparate health telemetry from Garmin, Fitbit, Oura, Apple HealthKit, and Terra into a unified schema. To resolve the complexity of heterogeneous data delivery—ranging from Garmin webhooks and Fitbit OAuth2 polling to raw HealthKit payloads—I developed provider-specific parsers and a centralized batch normalization engine to standardize metrics such as HRV, SpO2, and sleep stages. The infrastructure was built on a serverless AWS Lambda architecture using Mangum and FastAPI, utilizing Docker to isolate provider dependencies and implementing a dual-layer storage strategy with Firestore for patient-centric documents and S3 for raw archival.
<aside>
Stack: Python · FastAPI · AWS Lambda · AWS SAM · boto3 · S3 · Firestore · PostgreSQL · Garmin / Fitbit / Oura / Apple / Terra APIs
</aside>
Developed a real-time event synchronization architecture between a clinical web dashboard and an Electron-based patient desktop application, leveraging PostgreSQL LISTEN/NOTIFY to enable low-latency push notifications for calls and messaging. I engineered an interactive notification system using Electron’s IPC layer and integrated Azure Communication Services (ACS) for seamless video consultation. To improve the user experience, I built a custom Firestore-backed live transcription service that overlays real-time captions within the call interface. I also implemented a global state management strategy to ensure call status remained synchronized across the entire React application and designed a multi-tenant routing logic to enforce strict data isolation between different healthcare organizations, ensuring secure and private communication channels.
<aside>
Stack: PostgreSQL LISTEN/NOTIFY · Next.js · Electron · Azure Communication Services · Firebase Firestore · TypeScript · React
</aside>
Assisted in the development of core clinical features for a physician-facing web application, focusing on high-performance data visualization and secure session management. I engineered a comprehensive patient health dashboard using Recharts, designing custom visualizations for complex biometric data—including intraday blood glucose, stress levels, and "Body Battery" metrics—while optimizing rendering performance by disabling animations for large-scale datasets. On the infrastructure side, I refactored the authentication layer to use Firebase ID tokens integrated with Iron Session, implementing an auto-refresh mechanism to maintain session persistence during clinical shifts. To improve data architecture, I centralized organization resolution within facility-level documents, enabling seamless cross-facility patient visibility. Additionally, I integrated PostHog for product analytics, configuring the provider and page-view tracking to drive data-informed UI/UX iterations.
<aside>
Stack: Next.js · TypeScript · React · Firebase Auth + Admin SDK · Firestore · Iron Session · Recharts · Tailwind CSS · PostHog
</aside>
Business Technology Intern — ZS Associates
Jan 2026 – June 2026
Built a browser-based multi-agent platform that automates the full data engineering lifecycle — from raw business inputs to production-ready requirements documents, data models, DDL scripts, and test suites — powered by Claude Code CLI as the agent execution backbone. To bridge the gap between heterogeneous agent workflows and a live browser UI, I architected a streaming WebSocket pipeline where each agent run executes as an isolated asyncio subprocess, fanning out real-time events (text deltas, tool invocations, file writes) to multiple concurrent subscribers via per-client asyncio queues — decoupling agent execution from transport entirely. I designed a skill package architecture where each agent's instruction set, slash commands, phase-level skills, and Python tools are defined in a self-contained directory and symlinked into every run directory, enabling zero-copy instruction updates across all historical and future runs. The platform supports a dual storage strategy — a local filesystem backend for development and a Postgres + S3 backend for production — abstracted behind a common interface so the application layer remains storage-agnostic. To handle long-running agent sessions across nginx proxies, I implemented application-level heartbeating that periodically emits keepalive events to reset upstream read timeouts without relying on WebSocket ping/pong frames, which do not traverse the proxy layer reliably.
<aside>
Stack: Python · FastAPI · Claude Code CLI · WebSockets · asyncio · PostgreSQL · AWS S3 · React · Zustand · Docker · Nginx
</aside>
Army Institute of Technology, Pune Bachelor of Engineering [Computer Engineering] September 2022 – June 2026
GPA: 9.2