Full stack developer
India · abhishekvish0212@gmail.com
Building quietly — interfaces, systems, and small tools that feel good to use.
Vercel clone that deploys any public GitHub repo (Vite/React) to a live subdomain in seconds — isolated Docker builds, queued via Redis/BullMQ, served from R2.
• Built a full stack multi-tenant RAG platform with a React frontend and FastAPI backend, using PostgreSQL for user accounts and application data through SQLAlchemy ORM with Alembic migrations, and ChromaDB for vector storage. • Replaced naive fixed-size text splitting with semantic chunking using Chonkie and sentence-transformer embeddings, keeping retrieved chunks coherent and aligned with actual meaning instead of cut off mid-thought. • Designed multi-tenant data isolation with metadata-filtered vector queries, so each tenant only ever retrieves its own documents, with async FastAPI endpoints to handle concurrent requests cleanly. • Added input/output guardrails for prompt injection, PII, and content safety, plus semantic caching to cut down redundant LLM calls and a fallback model chain with exponential backoff and jitter for reliability. • Built per-request and aggregate cost tracking, structured logging with request IDs, and evaluation logging, so every LLM call stays traceable and easy to debug in production.
- Built a full-stack web application that enables users to track daily activities, habits, mood, health, career progress, and financial well-being through structured logs and journal entries. - Designed an AI-powered insight engine that analyzes user behavior over time and generates personalized summaries, trends, recommendations, and overall assessments. - Implemented background processing workflows to generate daily, weekly, and monthly behavioral insights, reducing real-time computation costs and improving response latency. - Architected a long-term memory system using hierarchical summaries, user profiles, and semantic retrieval to provide personalized AI responses based on historical user data. - Developed analytics pipelines to identify habit trends, correlations, streaks, and behavioral patterns across multiple life domains. - Integrated LLM APIs for contextual coaching, reflection generation, and natural language question answering over user history. - Built secure authentication and user management, deployed the application on Vercel, and configured custom domain management through Cloudflare. - Tech Stack: Next.js, TypeScript, PostgreSQL, OpenAI API, pgvector, Cloudflare, Vercel.
- Built an AI-powered Job Application Copilot using TypeScript and the OpenAI Responses API with a custom agentic tool loop - Implemented 4 specialized tools: job description analysis, PDF resume parsing, resume-to-job matching with scoring, and tailored cover letter generation - Designed a structured agent architecture with a tool registry and iterative reasoning loop that autonomously decides which tools to invoke based on user input - Used Zod schemas for structured output extraction from LLM responses, ensuring type-safe and validated data across all tool calls - Parses PDF resumes using pdf-parse and extracts structured job requirements from unstructured text via GPT-5.2 - Generates cover letters with configurable tone and length, outputting polished markdown files ready for submission - Handles raw terminal input with bracketed paste mode support for seamless multi-line job description pasting
- Automated job application tracker that monitors Gmail every 3 hours and logs application statuses to Google Sheets - Uses OpenAI's GPT-4o-mini to classify and extract structured data from emails — company, role, platform, and status — with JSON schema enforcement - Supports 16+ job platforms (LinkedIn, Greenhouse, Lever, Workday, Ashby, etc.) with domain-based fallback detection - Idempotent processing via SQLite state store ensures no duplicate entries across runs - Monotonic status progression logic prevents status regression (e.g., Rejected never reverts to Applied) - Runs on macOS (launchd), Linux (crontab), and cloud (GitHub Actions) with full CI/CD pipeline - Modular Python architecture — clean separation of Gmail client, Sheets client, LLM extraction, and state management - DRY_RUN mode for safe testing without writing to Sheets - OAuth2 token caching with auto-refresh for both Gmail and Google Sheets APIs - Privacy-conscious design — email bodies truncated to 6KB before sending to OpenAI, all secrets gitignored
- Built a self-custody Solana wallet and trading platform with live candlestick charts, real-time token discovery, and on-chain swap execution via Jupiter V2 aggregator — all running on Next.js 16 and React 19. - Integrated Privy for Google/Apple OAuth with auto-provisioned embedded Solana wallets, enabling users to sign transactions via a secure iframe without managing seed phrases. - Engineered a server-side API proxy architecture that keeps third-party keys (Codex, Jupiter, Alchemy) off the client while powering real-time price polling, OHLCV bars, holder analytics, and swap quoting/execution. - Built a professional-grade swap flow with debounced quote fetching, base64 transaction decoding, Privy signing, execution tracking, and Solscan confirmation links.
- Built a full-stack rental property marketplace using Next.js 14 (App Router) with server/client component architecture and Prisma/PostgreSQL for data modeling - Implemented AI-assisted listing creation via LLM text-to-structured-data parsing, reducing manual entry effort; built admin CRUD with inline editing, deletion, and inquiry management with CSV export - Developed a public-facing filterable listing grid with location, rent, type, and furnishing filters, WhatsApp contact integration, and inquiry submission tied to specific listings - Added class-based dark mode across 20+ components using a custom ThemeProvider with system-preference detection and localStorage persistence, styled with Tailwind slate palettes
- Self-contained, web app for learning system design from the ground up - Structures a full curriculum into five progressive phases: prerequisites, foundations, core patterns, advanced concepts, and real-world case studies - Interactive phase/topic navigation with modals, search, and progress tracking - Light/dark theme toggle with smooth transitions - Each topic rendered from Markdown content via a JSON-driven catalog - Built with vanilla HTML, CSS, and JavaScript — no backend or build step required
- Built a platform to monitor infrastructure and service health by integrating tools like Zabbix and NinjaOne into a centralized system. - Developed a Windows Service that continuously collects monitoring data from client environments and sends it to the platform for processing. - Designed a background status engine that evaluates component health and automatically determines overall service status based on dependency rules. - Implemented incident tracking for component outages and mapped affected services to improve root cause visibility and operational debugging. - Added support for maintenance scheduling, service impact analysis, and real-time status updates for monitored systems. - Improved API and database performance by optimizing queries, adding indexes, caching frequently accessed data, and reducing unnecessary database scans. - Performed load testing using k6 to test ingestion throughput, database load, and background processing performance. - Worked on deployment planning for both Azure cloud and VM/on-prem environments using: - Azure App Service - Azure Functions - Azure SQL Database - Built admin features for logs, traces, filtering, and observability to help track issues across services and components. - Focused on building a scalable and maintainable architecture by separating raw monitoring data, evaluation logic, and user-facing service status.
- Built a full-stack "Today I Learned" journaling application with AI-powered note formatting using Next.js 16 (App Router) and TypeScript - Integrated OpenAI GPT-4o-mini to transform raw, unstructured notes into clean, well-formatted Markdown entries - Implemented Google OAuth authentication with NextAuth v5 and session management - Developed a rich text editor using TipTap v3 with manual editing as an alternative to AI formatting - Designed a RESTful API with CRUD endpoints for journal entries, supporting full-text search, tag-based filtering, and date range queries - Styled the application with Tailwind CSS v4 and Typography plugin, including a system-aware dark mode toggle - Set up PostgreSQL 17 database with Prisma v7 ORM, including schema migrations and relation mappings for users, accounts, sessions, and journal entries - Containerized the application with a multi-stage Dockerfile (Node 22 Alpine) and Docker Compose for one-command deployment - Built a responsive sidebar UI for searching, filtering by tags/dates, and theme switching
- Built a full-stack portfolio showcase app with Next.js 16 (App Router), TypeScript, and Tailwind CSS v4 - Implemented secure admin dashboard with NextAuth v5 (JWT-based credentials authentication) and protected API routes - Designed a glass-morphism UI with dark/light theme toggle, gradient accents, and smooth Framer Motion animations - Created a drag-and-drop project reordering system using @hello-pangea/dnd with atomic database updates via Prisma transactions - Set up PostgreSQL with Prisma ORM using the driver adapter pattern for connection pooling and SSL support (Neon.tech compatible) - Built responsive layouts with mobile-first design — sidebar on desktop, bottom navigation on admin pages - Implemented file upload system with image preview and timestamped random filenames - Containerized with multi-stage Docker build (Node 22 Alpine) and Docker Compose with persistent volumes - Used Server Components for data-fetching pages and Client Components for interactive features (login, forms, DnD) - Designed a custom component library (Button, Input, Card, Badge) with a cn() utility for class merging - Managed project ordering with a dedicated reorder API endpoint using Prisma batch updates - Deployed with standalone output mode for optimized Docker images and optional schema push on startup
- Built a two-sided marketplace for financial talent (Accountants, CFOs, AR Specialists). - Implemented a complete **GraphQL API** (queries + mutations) for auth, profiles, requests, proposals, engagements, reviews, favorites, and notifications. - Added **natural-language expert search** using OpenAI GPT‑4o: unstructured query → structured filters (Zod validation). - Implemented role-based access control (Business/Expert) with JWT authentication and a quote-to-hire workflow with status tracking. - Containerized with docker and deployed on **AWS EC2** with **Caddy** reverse proxy (frontend static + backend API). - Used Neon (serverless PostgreSQL) for cloud-hosted database.
Identified SQL metadata exposure vulnerability in production
Jul 2021
Bachelor's · Computer Application
Jun 2019 — Jun 2022
A brief note is enough — I’ll reply within a day.
No spam — just a direct message.