hashkat.Ashutosh.
Research Engineer @ Adobe · LLM Agent Evals & Benchmarking · AI Security
SDR-Bench · Memento · CSAW ESC '22 winner · WACV & ECCV · Rust systems · IIT Roorkee
// What I work on:
const research = {
agents: ["LLM evals", "agent benchmarking", "RL fine-tuning"],
security: ["adversarial ML", "CTF", "Web3 security"],
systems: ["Rust", "OS kernels", "GPU clusters"],
infra: ["K8s", "vector DBs", "distributed systems"],
thesis: "measure what agents can't do yet"
};~/publications
Research in AI/ML and computer vision with collaborators from top institutions
4 publications across WACV, ECCV, and BiTW • Collaborations with Adobe Research, Stanford, Microsoft Research, CMU, and premier IITs
MEMENTO: Leveraging Web as a Learning Signal for Low-Data Domains
SDR-Bench: Benchmarking the Personalization Capabilities of Large Language Models
ReEdit: Multimodal Exemplar-Based Image Editing with Diffusion Models
Towards Efficient Exemplar Based Image Editing with Multimodal VLMs
~/achievements
Recognition for contributions to AI research, blockchain innovation, and cybersecurity competitions
CSAW ESC 2022 — 1st Place, Research Track
World's oldest hardware security competition · Adversarial attacks on ML models
Won the Embedded Security Challenge research track at NYU for adversarial attacks against machine learning models — before AI security became a mainstream research area.
NEAR REDACTED 2024 — 1st Prize
Web3 security CTF competition · Bangkok
Won 1st Prize at the NEAR REDACTED 2024 CTF in Bangkok, demonstrating expertise in Web3 security, smart contract vulnerabilities, and blockchain exploitation.
EigenLayer Infinite Prize
ValidAI — Decentralized AI validation via AVS
Won the EigenLayer Infinite Hackathon for ValidAI, an Actively Validated Service (AVS) leveraging EigenLayer's restaking infrastructure for decentralized AI model verification.
CTF Global Rankings — Team Captain
InfoSecIITR · Ranked #4 in India, #40 globally
Led InfoSecIITR as Team Captain to rank #40 globally and #4 in India on CTFtime. Built and managed infrastructure for BackdoorCTF and Hackentine competitions.
GIAC Foundational Cybersecurity Technologies
SANS Foundation · GFACT Certification
Holds the GIAC Foundational Cybersecurity Technologies (GFACT) certification from the SANS Foundation, validating core cybersecurity knowledge.
~/experience
Applied Research Engineer
Working on the Sales Qualifier Agent (AJO B2B AO) — an AI-driven application that automates B2B prospect qualification and outreach. Leading personalization research including SDR-Bench, the first benchmark for measuring personalization capabilities of Deep Research agents for B2B sales.
Research Intern
Built RadBot — an automated brain tumor segmentation and survivability prediction system. Applied multimodal AI to medical imaging at the intersection of ML and clinical neuroscience.
Infrastructure Engineer
Managed Kubernetes clusters and microservices across AWS and GCP. Built event-driven architectures with Kafka pipelines, Redis caching, and MySQL binlog processing for real-time data sync.
Research Intern
Developed ReEdit — a novel end-to-end framework for exemplar-based image editing using diffusion models and VLMs. Published at WACV 2025 and ECCV 2024 AI4VA workshop.
Team Captain
Led cybersecurity initiatives and CTF competitions (Rank #40 globally, #4 in India on CTFtime). Won CSAW ESC 2022 in Research Track for adversarial attacks against ML models. Mentored team members and developed security tools and frameworks. Visit: infoseciitr.in
Developer
Contributed to open-source projects at SDSLabs, IIT Roorkee. Developed VectorDB, Katana, RusticOS, and participated in multiple hackathons. Visit: sdslabs.co
~/expertise
Multi-domain technical expertise across cutting-edge technologies
Agentic AI & Evaluation
End-to-end architecture of multi-agent LLM systems. Pioneer of SDR-Arena and Memento benchmarks. Built RL fine-tuning pipelines on 10+8 x A100 GPU clusters with 1.5-3x inference speedups.
Multimodal & Vision-Language
ReEdit (WACV/ECCV) for exemplar-based image editing with diffusion models. RadBot for brain tumor segmentation. Knowledge graphs via internet-scale scraping for LLM traversal.
AI & Systems Security
CSAW ESC 1st place for adversarial ML. NEAR REDACTED 2024 CTF winner. Team Captain of InfoSecIITR (#4 India, #40 globally). GIAC/SANS certified.
Infrastructure & Distributed Systems
K8s microservices at Abacus.AI across AWS/GCP. Kafka/Redis/MySQL binlog pipelines. GPU orchestration for distributed RL training.
Low-Level Systems
RusticOS — modular kernel in Rust with ring 0/3 separation and custom syscall interface. VortexDB — vector database engine with HNSW indexing and DistilBERT vectorizer.
Blockchain & Web3
EigenLayer Infinite Prize winner (ValidAI AVS). Proof of Optima using RISC-0 zkVM and Solidity. MCP servers and A2A frameworks for production SSE serving.
~/projects
Open-source contributions and personal projects across multiple domains
SDR-Bench
AI/MLFeaturedThe first framework to systematically benchmark generative personalization capabilities of LLMs for B2B sales. Features a dual-layered dataset spanning 6,279 articles across 20+ industries.
RusticOS
OS DevelopmentModular operating system kernel written completely in Rust. Features custom memory management, process scheduling, and x86-64 architecture support.
VortexDB
AI/MLHigh-performance vector database core engine built from scratch in Rust. Features HNSW vector indices, DistilBERT image vectorizer flows, and efficient similarity search for ML applications.
RadBot
AI/MLAutomated brain tumor segmentation and survivability prediction system. Applies multimodal AI to clinical neuroimaging for accurate diagnosis support.
Katana CTF Platform
SecurityHighly available K8s attack-defense CTF platform with strict namespace isolation, automated MongoDB/MySQL provisioning, and health-check cronjobs.
ValidAI
BlockchainFeaturedDecentralized AI validation system leveraging Actively Validated Services (AVS). Implements custom consensus for ML model verification on-chain. Winner of EigenLayer Infinite Prize.
Proof of Optima
BlockchainZero-knowledge proof system for verifiable computation using zkVM and RISC-0. Demonstrates advanced cryptographic protocols for smart contracts.
~/skills
Technical expertise across multiple domains
AI & Machine Learning
Security & CTF
Infrastructure & DevOps
Blockchain & Web3
Languages & Tools
~/contact
Open to research collaborations, speaking opportunities, and exciting projects