AI EngineerAgentic systemsBengaluru, IN

Shriyans Kandhagatla

call me Shri

I'm here as

Ask my AI twin → Follow a task ↓

0 dotshover to scatter · click to drop

How this page works

Follow one task through an agent.

This page runs in the order my agents do. Scroll, and your visit is the task.

01Orchestrator

Who's handling your task.

I build agentic systems that hold up in production. Today that's AI features for HVAC estimators at Torinit. Claude Code is open on my screen all day.

2026.05 → now	Torinit	Software Engineer, AI · features for HVAC estimators
2024.06 → 2026.04	NStarX	Data Scientist / AI / ML Engineer
2024.01 → 05	Futurense	AI Data Engineer
2024.08	Jain University	B.Tech CSE (AI & ML) · CGPA 8.0

NStarX: where the curve bent upward

Joined before graduating. Left leading platform design. Tap a point.

02Tools

The tools I reach for.

drag to orbit · click a tool to call it

Certs: Anthropic Claude Code · Hugging Face MCP Fundamentals · Microsoft AI-900 · Simplilearn NLP · Microsoft & LinkedIn Generative AI · more coming soon

03Guardrails

What I take on.

ALLOWcore

Agentic systems and multi-agent setups
Agent tooling and MCP
LLM gateways and model routing
Forward deployed work with client teams

FLAGcan do

React and FastAPI features
Fine-tuningheld back by data and GPUs, not interest
Retrieval, now as agent memory

04Execute

One platform, split into parts anyone can plug in.

At NStarX I led four shared layers that turned a monolith into services. Mine are highlighted.

use cases

RAG Studio

Multimodal Chatbot

Report Generator

↓ consume ↓

shared layers

Interactionunified LLM gateway, streaming, rate limits, prompts

VectorStoreadapter pattern, backend picked per request

GuardrailsNeMo Guardrails, PII, toxicity

EvaluationRagas and TruLens on RAG quality

↓ route to ↓

backends

OpenAI · Anthropic · Groq · Gemini · NVIDIA NIM · vLLM · Ollama on H200 GPUs | Milvus · ChromaDB · Weaviate · Qdrant

Torinit

AI features for HVAC estimators

New features that help estimators work faster, across React and FastAPI.

NStarX

Multi-agent event planner

Peer-to-peer agents on NVIDIA NAT with SQL tools and structured output.

05Reflect

Figure it out, then make it hold.

RequirementsWhat counts as done.
Inputs and outputsPin the contract.
Failure pathsRetries, fallbacks, edge cases.
MeasureNumbers before opinions.

Patterns I build, and work by

The highlighted node is me.

pattern · routing

Routing

Every task gets triaged first: agent, workflow or plain script.

pattern · parallelization

Parallelization

Several projects at once, on separate tracks, merged at checkpoints.

pattern · evaluator-optimizer

Evaluator-optimizer

Draft, measure, refine, until the numbers hold.

ExploringJev by TypeSafe AIGame developmentAI for engine diagnostics

Off the clock

Where the hours go.

Pick a spot.

Contact

Ask, then reach me.

Ask my AI twin anything about my work. For anything that matters, email me.

twin@shriyans:~local notes · guardrails on

Tap the card for a QR code.