shivam honrao

●listening toMidnight City — M83📍 (पुणे, इंडिया)

AI ENGINEER · FULL-STACK · ARCHITECT

today

I'm an AI engineer and full-stack architect building AI-powered products — LLM apps, agentic systems, and the full-stack around them.

enjoys side projects, cricket and badminton, clean interfaces and learning new stacks.

i also follow and contribute to open source and write about what I learn while building agents.

model card

modelshivam-honrao · full stack ai engineer · agentic engineer · ai-native builder
architectureagentic loop · sandboxed execution · memory + context layer · LLM inference · real-time infra
training corpuscoding agents, voice AI, video AI, RL environments, RAG pipelines, multi-agent systems, loop engineering, eval harnesses, AI-native products
context windowAI-native spec → agentic loop → tool-call → memory write → eval harness → production
capabilitiesagentic engineering, coding agents, sandboxing + infra, voice + video AI, context engineering, memory systems, evaluation harnesses, loop engineering, multi-agent orchestration
evalsships AI-native products end to end — agents that plan, execute, reflect, and complete
known limitationscannot stop turning every problem into an agentic loop

past

Excello Circuits, Freelance Software Engineer

Kiwi Group, Founding Software Engineer

miniOrange, Software Engineer

Bluestock Fintech, Software Engineer Intern

personal projects

OpenDev preview

A terminal-based CLI AI coding assistant built from scratch, utilizing system tool-calling, terminal state persistence, sandboxed executions, and multi-agent task orchestration.

↗ code

mini apps ↗NEW

collection of small apps and games, sharing them here in case they're useful to you too.

ai playground ↗NEW

interactive demos that run in your browser — see how tokenizers split text and how embeddings measure meaning.

writing

"Beyond the ReAct Loop: Engineering Long-Horizon AI Systems with Deep Agents"28d ago

An architectural deep dive into shifting from simple reason-and-act loops to structured deep agents with persistent memory, orchestration, and human-in-the-loop controls.

"Building Intelligent Agents with LangChain: Runnables, Tools, and Human-in-the-Loop"31d ago

A deep dive into LCEL, custom tools, and implementing human oversight in autonomous agent workflows

all articles ↗ · more on medium ↗

research

Anti-Theft Mobile Tracking and Monitoring SystemIJRASET

freelance

open for freelance projects, let's build something together

✦⌘K