Open to software & AI roles

Suchith Koduru

Software engineer who ships production LLM agents, voice loops, and the backends that keep them honest.

Recently at 10xScale.ai — domain agents, edge voice AI, and real-time interview systems. B.Tech Computer Science, IIT Delhi ’25.

LLM agents Voice AI MCP / tools Backend systems
Say hello Resume
Suchith Koduru

01 About

less hype, more systems

I care about the boring parts that make AI products work: state, tools, latency, memory, and what happens when the model is wrong. At 10xScale.ai I owned slices of production agent systems — from FastAPI + SSE backends to voice pipelines on Raspberry Pi 5 and an enterprise voice interview loop.

I’ve contributed to the open-source agent framework AgentFlow (PRs #17 and #19), shipping Qdrant-backed long-term memory and hardening a GenAI adapter. Before that: B.Tech CS at IIT Delhi, plus research and ML work that taught me to measure before I mythologize.

Now

  • Looking for software / AI engineering roles with real ownership
  • Writing about agent systems, memory, and production tradeoffs
  • Still tinkering with voice loops, MCP tools, and edge inference

02 Selected work

case studies, not logo walls

Voice edge medical kiosk

10xScale

Async VAD → STT → agent → TTS on Raspberry Pi 5 with Indic language auto-detect and ~70% barge-in latency cut.

  • PySide6
  • FastAPI
  • Voice AI

AgentFlow

Open-source contributions: Qdrant long-term memory module and GenAI adapter edge-case hardening with regression tests.

↗ PR #17 ↗ PR #19
  • Python
  • Qdrant
  • Agents

Deep Research Agent

Autonomous web research agent for the Hugging Face MCP Hackathon — research, synthesize sources, write answers.

  • MCP
  • LangChain
  • Python
All projects & architecture notes →

03 Experience

what I owned, not buzzwords

Jan ’26 – Jun ’26

AI Software Engineer

10xScale.ai · Hyderabad

  • Built a domain medical agent on AgentFlow with skill playbooks, FastAPI + SSE streaming, checkpointer short-term state, and Qdrant long-term memory.
  • Owned end-to-end voice kiosk software on Raspberry Pi 5 (11 Indic languages) and cut barge-in latency ~70% (1000 ms → 300 ms).
  • Owned AI interviewer logic and sub-700 ms voice turns for an enterprise interview platform; coding-round evaluation APIs and structured reports.

Jun ’25 – Dec ’25

AI Software Engineer Intern

10xScale.ai · Hyderabad

  • Shipped a FastMCP server for PostgreSQL NL→SQL and Typesense search, wired into a multi-agent database node with auth middleware.
  • Merged AgentFlow PRs #17 and #19 — Qdrant long-term memory and Google GenAI adapter hardening.

2021 – 2025

B.Tech, Computer Science & Engineering

Indian Institute of Technology Delhi

  • ML coursework and projects including CIFAR-100 calibration with WideResNet + SAM, and earlier maternal health risk modeling.
  • JEE Advanced 2021: AIR 479 · JEE Mains 2021: AIR 620

04 Writing

notes from the trenches

Systems notebook Interactive

How Do You Stop Two People From Booking the Same Seat?

Race conditions, row-level locking, seat holds, payment outside transactions, and idempotent booking APIs — as an interactive walkthrough.

05 Toolkit

Day-to-day: Python, FastAPI, async streaming (WebSockets / SSE), AgentFlow / LangGraph, MCP, Qdrant, PostgreSQL, and enough React / Electron / PySide6 to ship the client side.

Python TypeScript FastAPI AgentFlow LangGraph MCP Qdrant Docker PyTorch

Open to interesting problems.

Prefer email over forms. If you’re hiring for software or AI systems work — agents, voice, backends, production ML — say hello.

suchithkoduru@gmail.com Resume PDF GitHub