Venkata Midhun Mareedu

Full stack developer and aspiring AI engineer in Hyderabad. I build LLM products with hard limits on themC, and I take closed software apart to rebuild itAB.

Seven exhibits on the table. Pick one up.

  1. PocketExpense tracker rebuilt from an APK
  2. Realme Link DesktopLinux app for Realme earbuds
  3. HireKitResume screening where every score cites a quote
  4. SynapseLocal AI PDF viewer
  5. ArchonAutonomous AI VC analyst
  6. Support AI ChatbotRAG support widget plus Go service
  7. OvermindLaTeX resume workflow

Exhibit A: Pocket

Android expense tracker, reverse-engineered

Findings
  • Rebuilt from MyMoney by analysing the shipped Android app.
  • Added UPI payment detection and an AI chat over your spending.
Made with
Kotlin, Android, REA, Claude Code
Open it

Exhibit B: Realme Link Desktop

Linux app for Realme earbuds, on a reverse-engineered Bluetooth protocol

Findings
  • A Linux replacement for the Realme Link phone app that controls Realme earbuds.
  • Talks to the device over a Bluetooth protocol recovered by reverse engineering.
Made with
Tauri, Rust, TypeScript
Open it

Exhibit C: HireKit

Resume screening that cites its evidence

Findings
  • Anonymized, evidence-backed screening: every score cites a verified quote.
  • All LLM calls pass through one gateway with token caps, logging and a USD 8 spend limit.
  • API, worker and web app run on a single Oracle Cloud VM with Docker, Caddy and Compose.
Made with
Python, FastAPI, PostgreSQL, React, Vite, Docker, Oracle Cloud
Open it

Exhibit D: Synapse

Local AI PDF viewer for the desktop

Findings
  • Desktop PDF viewer with inline question answering and semantic search.
  • Answers send you to the page they came from.
  • Embeddings live in LanceDB for low-latency retrieval, with models served by Ollama.
Made with
Electron.js, React, LanceDB, Ollama, LangChain.js, Vercel AI SDK
Open it

Exhibit E: Archon

Autonomous AI analyst for venture capital

Findings
  • Built the founders portal, memo approvals, spreadsheet import and deal scoring.
  • Added an email deal inbox, Telegram deal chat and bring-your-own LLM keys.
  • Portfolio reminders that draft the LP newsletter.
Made with
Python, FastAPI, Supabase, React, TanStack Query, Expo, Docker, Terraform
Held at
DeeptaAI (company work, source not public)

Exhibit F: Support AI Chatbot

RAG support widget and its Go service

Findings
  • A support chatbot shipped as an NPM widget and a Go server.
  • Retrieval-augmented answers with configurable prompts and pluggable tools.
Made with
React, Go, MongoDB, ChromaDB, Vercel AI SDK
Open it

Exhibit G: Overmind

LaTeX resume workflow

Findings
  • Produces ATS-tuned, role-specific resume variants from one source.
  • Reproducible PDF output every time.
Made with
Skill.md, LaTeX
Open it

Chain of custody

Who I have worked for, when, and what I did there.

  1. Aug 2025 to Present

    DeeptaAI

    SDE-I, Hyderabad

    • Architected AI Chatbot and Form SDKs in a Turborepo for cross-platform university integration, with reusable LLM orchestration patterns.
    • Engineered academic infrastructure modules for complex program and course hierarchies.
    • Implemented secure multi-channel payments for self-funded and loan-based transactions.
    • See exhibitE.
  2. Dec 2024 to Jul 2025

    Apxor

    Associate SDE, Hyderabad

    • Built a generative AI chatbot that turns natural language into ClickHouse SQL for automated analytics and visualization.
    • Optimized web performance to 98% with caching strategies, Lighthouse audits and monitoring.
    • Led frontend development for high-traffic, AI-assisted analytics platforms.
  3. Feb 2024 to Aug 2024

    IncrivelSoft, Inc

    Full Stack Web Developer, Hyderabad

    • Co-led the frontend team delivering responsive MERN interfaces and high-fidelity UI components.
    • Integrated complex UI with backend APIs for reliable data flow end to end.

Tools on the bench

Languages
Python, Java, JavaScript, Go
AI
Generative AI, LLMs, RAG, prompt and context engineering, LangChain.js, Vercel AI SDK, Ollama, vector databases, MCP servers and clients
Reverse engineering
REA, APK analysis, protocol reverse engineering, Claude Code skills, Cursor, DeepSeek
Frameworks
React.js, Next.js, Electron.js, Node.js, Express.js, FastAPI
Tooling
Turborepo, Vite, Docker, Kubernetes, Git, GitHub Actions, GitLab CI/CD, Vitest, Playwright
Cloud
AWS EC2, Google Compute Engine, Oracle Cloud VMs, Terraform
Interface
Tailwind CSS, shadcn UI, Framer Motion, TanStack Query and Router, Zustand, React Hook Form, Zod
Databases
MySQL, PostgreSQL, MongoDB, ChromaDB, LanceDB
Education
B.Tech, Computer Science and Engineering, JNTU-GV, 2021 to 2024

Sign one out

Every exhibit is open for inspection. If your team needs someone who ships LLM products and checks the receipts, write to me.

midhunmareedu@gmail.com

Venkata Midhun Mareedu, Hyderabad, India. Models on this table are built in code with three.js. Back to the top