PRAKHARAI SYSTEMS THAT DEFEND PEOPLE
PS©2026

Prakhar Shukla, software engineer of applied AI systems

Software engineer. Applied AI systems.

I build AI systems that defend people: firewalls for hallucinations, honeypots for scammers, graphs for fraud rings. Shipped, measured, honest.

Resume.pdf
AWS AIdeas Global Top 50Top 200 AI teams in IndiaIEEE + journal publicationsNagpur, India

Trust is an engineering problem.

I build systems that earn it:

hallucination firewalls, scam honeypots,

deepfake detectors, fraud graph engines.

Every number here regenerates from code.

Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring · Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring · Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring · Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring · Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring · Hallucination firewalls · Scam honeypots · Deepfake detectors · Fraud graph engines · Evidence bundles · Deterministic scoring ·

01 Selected work

Five systems that catch what slips past everyone else.

01

Sentinel

Cross-merchant fraud defense · 2026

Cross-merchant UPI fraud detection. An identity link graph catches recycled devices and VPAs between merchants, scores every event deterministically, and hands analysts evidence instead of a black box number.

0.857
F1, held-out set
2.3 ms
p50 scoring latency
+₹38,665
net saving per 1,000 events
91%
test coverage, 160 tests
PythonFastAPISQLiteAWS BedrockNext.js
02

Gatehouse

Autonomous fraud-defense agent · 2026

A household forwards any suspicious message to a team of AI agents that investigates it like a professional fraud analyst and escalates only genuine threats with court-grade evidence bundles.

20
controlled design documents
30-case
offline evaluation harness
0
network calls in the rule engine
Wilson
intervals on every metric
Strands Agents SDKAmazon Bedrock AgentCorePythonPydantic
03

SignalGate

Agentic research-integrity gate · 2026

Candidate quant trading signals investigated like fraud cases. Statistical probes run as sandboxed tools, verdicts come with numeric receipts, and spurious signals die before they waste a researcher's hour. Recommends, never trades.

0.925
spurious catch vs 0.475 baseline
p=0.00029
McNemar paired test
0.0
false-reject on sound signals
Byte-identical
eval repro, asserted in CI
PythonFastAPITyperHTMXTailwind
04

TruthLayer

AWS AIdeas, Global Top 50 of 10,000+ · 2026

A serverless firewall that intercepts LLM hallucinations before users see them. Five deterministic signals, 95.33% precision on 300 adversarial cases, $1.50 a month at 50K verifications.

Top 50
global finalist, AWS AIdeas
95.33%
precision on adversarial cases
$1.50/mo
at 50K verifications
5-signal
deterministic engine
Amazon BedrockLambdaDynamoDBAPI GatewayTypeScript
05

ScamShield

India AI Impact Buildathon, Top 2% of 40,000+ · 2026

Autonomous agents engage scammers with adaptive personas, extract UPI IDs and bank details with 98% accuracy, and feed a live intelligence dashboard over 13 REST APIs.

98%
intelligence extraction accuracy
13+
REST APIs on the dashboard
Adaptive
LLM persona switching
FastAPIGROQMongoDBNext.jsRender

02 Research

8th IEEE Pune Section International Conference, PUNECON 2025

Comprehensive Deepfake Detection: A Comparative Study on Image and Video Forgery

i-manager's Journal on Image Processing

Advanced Deepfake Image Detection: A Robust Framework using InceptionV3 and Xception

03 Writing

04 Recognition

AWS AIdeas Global Top 50 Finalist

AWS AIdeas Global Top 50 Finalist

TruthLayer, picked from 10,000+ global submissions. Published by AWS Builder.

Proof ↗
Top 200 AI Teams in India

Top 200 AI Teams in India

India AI Impact Buildathon, grand finale at Bharat Mandapam, New Delhi. Top 2% of 40,000+ builders.

National Winner, SBI Youth Ideathon

National Winner, SBI Youth Ideathon

Top 15 of 15,000+ teams at IIT Delhi. ₹20,000 for Truth Shield.

Proof ↗
National Innovation Finalist

National Innovation Finalist

BITS Pilani Beyond Profits Top 20 and Azim Premji Social Enterprise Challenge finalist.

1st Runner-Up, Hack Wack 2.0

1st Runner-Up, Hack Wack 2.0

2nd of 300+. ₹8,000 for real-time deepfake detection.

1st Runner-Up, HackAIthon

1st Runner-Up, HackAIthon

Azure Developer Community and Reskilll, applied AI track.

05 Journey

2022

Started B.Tech in CSE, RTM Nagpur University.

2023

First shipped project: the school website, maintained for a year.

2024

Went deep on deepfakes. First detector, then first paper.

2025

IIT Delhi win, IEEE paper, journal paper, internship at Anishree Technologies.

2026

Graduated at CGPA 8.79. AWS Global Top 50. Top 200 in India at Bharat Mandapam.

06 Lab

Things built because they had to exist.

DRAG THE MERCURY

07 Contact

Build something
worth defending.

prakhar230125@gmail.com

Open to full-time roles and hard problems in applied AI, trust and safety, and systems that ship.