# Piyush Choudhari - Full Content

> AI & Backend Engineer specialized in Agentic Systems, LLM Evaluation, and High-Performance Backend Design.

## Blog Posts

- [Building Mentis: A Memory Layer for Coding Agents](https://piyushc.com/blog/Building-Mentis-A-Memory-Layer-For-Coding-Agents) (2026-09-26)
  Summary: How Mentis stores coding-agent attempts, retrieves related tasks, and marks outdated conclusions.

- [Sandboxing Coding Agents](https://piyushc.com/blog/Sandboxing-Coding-Agents) (2026-09-03)
  Summary: Building a coding harness and a sandboxing platform to create self-driving development.

- [Making Agents Play Pictionary](https://piyushc.com/blog/Making-Agents-Play-Pictionary) (2026-05-17)
  Summary: A turn-based drawing-and-guessing game where AI agents compete

- [My Codebases Have an AI Receptionist Now](https://piyushc.com/blog/My-Codebases-Have-an-AI-Receptionist-Now) (2026-04-08)
  Summary: I built an agentic codebase navigator that lets engineers and recruiters ask anything about my projects, no RAG, no embeddings, just pure read-and-reason.

- [Implementing An SQS Like Message Queue System](https://piyushc.com/blog/Implementing-An-SQS-Like-Message-Queue-System) (2026-01-08)
  Summary: I built a rudimentary AWS SQS clone with managed queues, visibility timeouts, DLQs.

- [How I Used PostgreSQL, Groq & Vercel's AI SDK To Create A Chatbot For My Portfolio Website](https://piyushc.com/blog/How-I-Used-Postgres-Groq-And-AI-SDK-To-Create-My-Portfolio-Website-Chatbot) (2025-12-25)
  Summary: My portfolio website contains a decent amount of text data, so I implemented a chatbot to interface with it.

- [Designing a Deterministic, Low Latency Decision Engine with C++](https://piyushc.com/blog/Designing-a-Deterministic-Low-Latency-Decision-Engine-with-CPP) (2025-12-19)
  Summary: A high performance, deterministic decision tree engine built in C++ that evaluates rules via gRPC with low latency, hot reloadable trees, and production ready observability.

- [Simulating vLLM's PagedAttention](https://piyushc.com/blog/Simulating-vLLMs-PagedAttention) (2025-12-08)
  Summary: An implementation of a simulation of PagedAttention in Python

- [Making A Peer Review System for My Blogs Using Google-ADK & Mem0](https://piyushc.com/blog/Making-A-Peer-Review-System-for-My-Blogs-Using-Google-ADK-And-Mem0) (2025-11-26)
  Summary: I needed an automation peer review my blogs, so I used Google-ADK and Mem0 to create an end to end system.

- [Training a Mixture-of-Experts Router](https://piyushc.com/blog/Training-a-Mixture-of-Experts-Router) (2025-11-17)
  Summary: I was curious about MoEs, so I implemented and tested one.

- [Building a Vector Database from Scratch - CapybaraDB](https://piyushc.com/blog/Building-A-Vector-Database-from-Scratch-CapybaraDB) (2025-11-04)
  Summary: A custom implementation of a toy vector database

- [Be Curious About Your Compute](https://piyushc.com/blog/Be-Curious-About-Your-Compute) (2025-09-24)
  Summary: After facing a blocker related to hardware, I decided to deep dive into it: An Explainer

- [How I Built an Automated Social Media Workflow with LangGraph](https://piyushc.com/blog/How-I-Automated-My-Social-Media-Workflow-with-LangGraph) (2025-09-14)
  Summary: For me, posting on social media is a tedious task and repetitive task: draft → edit → get feedback → post. So I used Langgraph and some AI magic sauce to automate away these tedious jobs.

- [SHAP values for GBTs – intuition + how they work internally](https://piyushc.com/blog/SHAP-values-for-GBTs) (2025-09-01)
  Summary: This blog explores how Gradient Boosted Trees can be explained using PDP, ICE, and SHAP — with TreeSHAP making black-box models transparent and highly interpretable.

- [Welcome to My Blog](https://piyushc.com/blog/welcome-to-my-blog) (2025-08-29)
  Summary: Introducing my new blog where I'll share insights about software engineering, AI, and technology.

## Projects

- [Agent Sandboxing](https://piyushc.com/projects/agent-sandboxing) (2026)
  A cloud coding-agent system that isolates repository work in Docker sandboxes and opens pull requests from agent-authored changes.
  Stack: TypeScript, Node.js, Docker, PostgreSQL, Prisma, Oauth.
  Outcome: Separates the agent control plane from the Docker execution plane: the agent never enters the sandbox, and sandboxes receive only short-lived, repository-scoped GitHub credentials.

- [CapyNodes](https://piyushc.com/projects/capynodes) (2026)
  An interactive system design canvas where an AI Judge scores your architecture for scalability, reliability, and failure modes.
  Stack: Python, Langchain, Groq, Django, PostgreSQL, LLMs.
  Outcome: CapyNodes features a unique multi-stage evaluation pipeline combining rule-based heuristics with LLM Chain-of-Thought (CoT) reasoning, backed by a Django + Channels architecture for real-time diagram state sync.
  - Active users: 50+
  - Evaluation sessions: 250+
  - Eval pipeline: 3-stage

- [CapybaraDB](https://piyushc.com/projects/capybaradb) (2025)
  A vector database built from scratch in Python, implementing indexing, similarity search, persistence, and the full query pipeline.
  Stack: Python, Pytorch, RAG, Vector Search.
  Outcome: Built from scratch using only numpy and PyTorch for core operations, providing complete transparency into vector database internals. Implements efficient exact search with linear scaling and sub-10ms query latency even at 5000+ vectors.
  - Query latency: <10ms
  - Recall @ k=5: 100%

## Experience

### AI Builder Intern @ Flytbase (Apr 2026 - Aug 2026)

- **FB Copilot**
  Built an AI assistant piloted across 10+ organizations for querying live drone operations and creating mission workflows through natural language, reducing setup time from ~5 minutes to under 30 seconds
  Designed a 6-stage planning pipeline for intent classification, parameter extraction, entity resolution, validation, planning, and operator confirmation before execution
  Built an 8-tool ReAct assistant using LangChain and Qdrant to answer grounded questions about fleets, telemetry, docks, missions, and alarms in under 5 seconds
  Developed a 300-case evaluation suite covering routing, parameter extraction, entity resolution, and rejection of unsupported or fabricated requests

- **Thermal Radiometric Inspection Tool**
  Built a DJI radiometric inspection platform with a backend that converts R-JPEG images into calibrated 640×512 temperature matrices in 150–300 ms, supporting per-pixel, region, line, histogram, isotherm, and image-comparison

### AI Engineer Intern @ Ronin Labs (Jan 2025 - Apr 2026)

- **AskRaven**
  Built an AI system that analyzes live Meta Ads data to answer performance questions, audit creatives, and recommend budget and creative actions. Used across 5 brands, resulting in campaigns with 4% average ROAS improvement
  Developed a two-layer personalization system combining researched Brand DNA with performance-grounded memory, and consolidated 22 specialized tools into 15 parameterized tools

- **Sketchbot**
  Built an image-to-sketch system deployed at 10 events and used by 1,000+ people, using FastAPI, SDXL Turbo, ControlNet, and OpenCV to cut robot drawing time from 10 minutes to under 3 minutes
  Deployed GPU inference on AWS g5.2xlarge instances, reducing infrastructure cost by 30% per region, with automated AMI/EBS recovery in under 15 minutes

- **OnePlus 13s Quest Quiz**
  Built backend APIs for a global OnePlus campaign serving 150K+ total users and 250+ active concurrent users, maintaining response times below 120 ms with Node.js, TypeScript, TypeORM, and GCP Cloud SQL

- **Cefaly**
  Built and optimized the pose-estimation system for Cefaly’s mobile app with 1K+ downloads, achieving 0.992 mAP50 for electrode detection and ~60 ms end-to-end mobile inference latency