Anuansh Tiwari

I research how generative AI breaks — and build tools that don’t.

anuanshtiwari191@gmail.com GitHub LinkedIn ORCID +91 98715 03105

Socrates · c. 470–399 BC

The unexamined life is not worth living.

Plato, Apology 38a

Measurement first

Claims are cheap. I build the corpus, run the experiment, and publish the numbers.

I work on the failure surface of generative AI: where safety filters give way, where agentic systems inherit vulnerabilities, where synthetic media erodes the assumption that evidence is evidence.

My method is first-party and reproducible. Survey claims get grounded in experiments I run myself — an n = 880 prompt corpus probing safety-filter behaviour, CVE data verified against the NVD rather than recycled from secondary write-ups, and figure pipelines in matplotlib that regenerate every plot from the source data so a paper's prose and its charts cannot drift apart.

The tools I ship follow the same instinct. In the Offline PDF Toolkit and GlucoLog Pro, nothing leaves the browser — not because a policy page says so, but because there is no server to send it to. Privacy as architecture, not as a promise.

0 PDF tools running client-side
Offline PDF Toolkit
0 prompts in the safety-filter corpus
first-party experiment
0 sources reviewed
resume-screening review
0 studies in the coded corpus
resume-screening review

Five papers

Research & publications

Surveys, reviews and measurements on how generative systems fail.

  1. P 01

    Weaponizing Generative AI: A Survey of LLM-Enabled Cyberattacks in Phishing, Malware Generation, and Social Engineering

    Published JIEEE vol. 07, no. 03 · Sep 2026 · presented at ICETA 2026, NITRA Technical Campus, Ghaziabad, 17–18 July 2026

    A survey of offensive uses of large language models across phishing, malware generation and social engineering, anchored by a first-party safety-filter experiment over an n = 880 prompt corpus.

    • LLM abuse
    • Phishing
    • Malware generation
    • Safety filters
    • n=880 corpus

    doi:10.54060/a2zjournals.jieee.212

  2. P 02

    Explainable Artificial Intelligence (XAI): A Systematic Literature Review of Methods, Evaluation Frameworks, and Applications in High-Stakes Decision-Making Systems

    Published JIEEE vol. 07, no. 03 · Sep 2026

    A PRISMA 2020 systematic review screening 9,140 records down to a 62-study corpus, covering explainability methods, how they are evaluated, and where they are applied in high-stakes decision-making.

    • XAI
    • PRISMA 2020
    • 9,140 records
    • 62 studies
    • High-stakes decisions

    doi:10.54060/a2zjournals.jieee.211

  3. P 03

    Agentic AI Security, Reliability and Trust

    Copyright transfer submitted IEEE review · four authors

    A review of the security, reliability and trust surface of agentic AI systems, built on NVD-verified CVE data and measured filter-performance results rather than secondary summaries.

    • Agentic AI
    • Threat modelling
    • NVD / CVE
    • Reliability

    TODO: add DOI / preprint link

  4. P 04

    AI-Based Resume Screening Systems: A Critical Review of Techniques, Failure Cases, and Industry Adoption (2024–2026)

    Conference talk · 8 min Springer format

    A PRISMA-informed critical review over 75 sources, narrowed to a 51-study coded corpus, covering screening techniques, documented failure cases and the state of industry adoption.

    • PRISMA-informed
    • 75 sources
    • 51 coded studies
    • Hiring algorithms
    • Springer

    TODO: add DOI / proceedings link

  5. P 05

    Post-Quantum TLS 1.3 Handshake Performance

    In progress Benchmarking against classical TLS 1.3

    Benchmarking ML-KEM hybrid key exchange against classical TLS 1.3 handshakes.

    • Post-quantum
    • ML-KEM
    • TLS 1.3
    • Benchmarking

    TODO: add results / preprint link when the work lands

Four shipped projects

Projects

Offline-first tools where privacy is architectural, not a promise.

ACM BuildHub

The technical-community platform for the ACM chapter at Amity Lucknow, built around a problems → projects → people → contributions loop: students pick a campus problem, form a team, and keep a record of what they built. An admin CMS edits every public page live with no redeploy, alongside a submission queue with review states, secured with Supabase row-level security, server-only keys and signed-cookie auth. A 96-assertion API test suite runs against an in-memory Supabase mock, with typecheck and build, on every push.

  • Next.js 15
  • React 19
  • TypeScript
  • Supabase
  • Row-level security
  • GitHub Actions CI

Live site Repository

Offline PDF Toolkit

53 PDF tools running entirely client-side. No server, no upload — the document never leaves the browser.

  • JavaScript
  • Client-side
  • Offline-first
  • Zero upload
  • AES-256

Repository

Offline PDF Toolkit — the Organize / Merge tool, with the full list of tools in the sidebar and the header stating it runs entirely in the browser Offline PDF Toolkit — the Lock (encrypt) tool: password-protect a PDF with AES-256 and set printing, copying, editing and annotation permissions

GlucoLog Pro

An offline single-HTML health tracker with a CGM screenshot digitizer, Whoop integration and Web Bluetooth device support.

  • Single-HTML
  • Offline
  • Web Bluetooth
  • Whoop API
  • Image digitizer

Repository

GlucoLog Pro — the home dashboard showing a 142 mg/dL reading in range, seven-day average, estimated A1C, variability and insulin-on-board tiles GlucoLog Pro — the Care and tools screen: care-team and emergency contacts, medical notes and an insulin sensitivity and carb ratio calculator

Amity Admissions Assistant

A Node.js / Express chatbot backed by Google Gemini, handling prospective-student queries through a single-page client. A voice-callback channel on ElevenLabs and Twilio was validated on free tiers first, alongside a university funding proposal covering the feature set and projected per-call API cost.

  • Node.js
  • Express
  • Gemini API
  • ElevenLabs
  • Twilio

Live demo Repository

Amity Admissions Assistant — a chat exchange answering B.Tech eligibility and the list of programmes offered, with suggested follow-up questions

Four roles, 2026

Experience

Telecom security, research and field programmes.

  1. Jun–Sep 2026

    Industrial Trainee — BSNL ALTTC, Ghaziabad

    Industrial training in Integrated Telecom / Data Networks and Cyber Security. Earlier produced a project report on security aspects in a smart building (Mar–Apr 2026), covering LAN and server-room infrastructure.

  2. Jun–Jul 2026

    Research Intern (NTCC) — Amity University, Lucknow

    Built and evaluated an AI resume-screening pipeline end to end — EDA, data cleaning, NLP preprocessing, TF-IDF feature extraction, then model training and evaluation — and wrote the accompanying critical review, under Prof. (Dr.) Wajahat Rizvi.

  3. Jun–Jul 2026

    HVCO Intern — Yuva Yoddha, Lucknow

    A five-week legal-rights awareness campaign: posters and an animated explainer, a 10-question survey in six languages, an offline community campaign, and analysis of the responses. Certificate of Appreciation from the founder.

  4. Jun–Sep 2026

    Teaching Intern — Each One Feed One, New Delhi

    Taught students in an education-access programme run by the NGO, alongside concurrent research and training commitments. Certified on completion.

Community

  • ACM Student Chapter — chartering effort

    Leading the effort to charter an ACM Student Chapter at Amity University, Lucknow, and built the chapter’s platform, BuildHub.

  • IEEE Student Member

    Member of the IEEE student community.

  • Research collaboration

    Recurring co-author with peers across institutions.

Skills

Languages, data tooling, model APIs and the security work they serve.

Languages

  • Python
  • JavaScript (ES6+)
  • TypeScript
  • SQL
  • C
  • HTML5
  • CSS3
  • LaTeX

Web & backend

  • Next.js
  • React
  • Tailwind CSS
  • Node.js
  • Express
  • REST APIs
  • Supabase
  • Streamlit

Data & ML

  • pandas
  • NumPy
  • scikit-learn
  • matplotlib
  • TF-IDF
  • NLP preprocessing
  • Model evaluation

AI tooling

  • Claude API
  • Gemini API
  • ElevenLabs
  • Twilio Voice
  • Prompt engineering
  • Claude Code
  • Cursor

Tooling

  • Git
  • GitHub Actions CI
  • Vercel
  • Postman
  • VS Code

Security research

  • LLM & prompt-injection security
  • Agentic-AI threat modelling
  • CVE / NVD analysis
  • ML-KEM
  • TLS 1.3
  • Network & smart-building security

Contact

Let’s compare notes.

Open to research collaboration, conference conversations and internship opportunities in AI security and applied cryptography.

anuanshtiwari191@gmail.com +91 98715 03105

Résumé — PDF