---
title: "AI / Agentic Engineer — Shivanshu Tiwari"
description: "Scam-detection agents, RL benchmarking systems, voice intelligence — built to run without a human in the loop, and documented honestly when they don't."
canonical: "https://shivanshutiwari.in/ai-engineer"
lastUpdated: "2026-09-02"
---

# AI / Agentic Engineer — Shivanshu Tiwari

I build autonomous agents and LLM pipelines

Scam-detection agents, RL benchmarking systems, voice intelligence — built to run without a human in the loop, and documented honestly when they don't.

## About

I don't just use AI — I build things with it that keep running after I close my laptop. I'm a third-year B.Tech CS & IT student in Bhopal, and over the past year I've built autonomous agents, LLM pipelines, voice intelligence tools, and RL benchmarking systems across 19 projects — some shipped and deployed, some deliberately stopped short of production so I could document what actually worked and what didn't. I'm not interested in demos that only look good in a pitch. When something fails — an RL agent losing to a simple heuristic, a classifier scoring 25% instead of the 90% I hoped for — I keep the result and figure out why, instead of reframing it until it sounds better. That's the standard I hold my own work to, and it's the standard I expect from anything I ship.

## Projects

- [DAITFO — RL Traffic Signal Control Benchmark](https://shivanshutiwari.in/projects/daitfo): A custom Gymnasium environment wrapping SUMO that benchmarks a PPO agent against fixed-time and queue-based heuristic controllers across 27 controlled runs. The heuristic won every metric — and that honest negative result is the actual finding, not a failure.
- [Agentic Honeypot — AI Scam-Baiting System](https://shivanshutiwari.in/projects/agentic-honey-pot): An AI agent that poses as a confused elderly retiree, engages real scammers in live conversation, and extracts UPI IDs, bank accounts, and phishing links from their own messages — turning reactive defense into active intelligence gathering.
- [Samvad — AI Meeting Assistant](https://shivanshutiwari.in/projects/samvad): Records in-person meetings, transcribes them live, and turns them into assigned, emailed tasks — without anyone taking notes.
- [CloudAuditEnv — LLM Agent Evaluation for Cloud Security](https://shivanshutiwari.in/projects/cloud-audit-env): Procedurally generated cloud security audit environment that tests LLM agents on finding and fixing real vulnerability patterns — built for the OpenENV MetaHackathon.
- [EchoPay — Audio-Powered Offline Payment System](https://shivanshutiwari.in/projects/echopay): I built an offline P2P payment system that encodes transaction data as audio tones between phones (MFSK modulation) — no internet required — with a hybrid UDP fallback and an offline-first Room sync queue.
- [Multilingual Mandi Platform](https://shivanshutiwari.in/projects/multilingual-mandi-platform): I built a real-time marketplace connecting local vendors across language barriers, with a provider-chain translation fallback (Google → Microsoft → AWS → mock) and Redis-backed WebSocket negotiation.
- [CyberMentor — AI Cybersecurity Companion](https://shivanshutiwari.in/projects/cyber-mentor): I built an AI cybersecurity companion for students — it scans suspicious links/emails and teaches through a teach-back quiz.
- [ComplianceIQ — AI Policy Compliance Scanner](https://shivanshutiwari.in/projects/compliance-iq): I built a tool where you upload a policy PDF, AI extracts the enforceable rules, and a rule engine scans live data for violations with human-in-the-loop review.
- [Saathi Community Assistant](https://shivanshutiwari.in/projects/saathi-community-assistant): I built a voice assistant that works in 13 Indian languages to help rural citizens find government welfare schemes.
- [Call Center Compliance API](https://shivanshutiwari.in/projects/call-center-compliance-api): I built an API that takes call-center audio, transcribes it, scores SOP compliance, and flags payment-intent signals — designed specifically to avoid the numeric hallucination problem that plagues pure LLM scoring.
- [AI Email Categorization Agent](https://shivanshutiwari.in/projects/email-categorization-agent): Local-first inbox automation agent — reads Gmail via OAuth2, semantically classifies every email into custom labels using an LLM, applies them automatically, and runs on a schedule with zero manual input. Skips already-labeled threads to avoid redundant processing.

---
*Source: https://shivanshutiwari.in/ai-engineer*