HARMFUL RESPONSES / −95% RESPONSE ACCURACY / +45% ENGAGEMENT / +30%
PHILOSOPHY / SAFETY-FIRST ENGINEERING
FRAME 02 — THE THESIS
TRAJECTORIES / 9,213 MONITORING COST / −75%
U. ARIZONA / ICML 2026 SUBMISSION
FRAME 03 — RESEARCH
ROLES / 04 SHIPPED PIPELINE GAIN / +20%
HANDSHAKE · ESCAPE · OMDENA · AMAZON
FRAME 04 — EXPERIENCE
CORE / PYTORCH · AWS · RAG DOMAINS / RL · LLM · NLP
TECH STACK / FULL SPECTRUM
FRAME 05 — SKILLS
DEGREES / 02 CERTIFICATIONS / 06
U. ARIZONA · VTU · CREDENTIALS
FRAME 06 — EDUCATION
STATUS / OPEN TO ROLES REMOTE & RELOCATION / YES
BENGALURU, IN / GLOBAL
FRAME 07 — CONTACT
HEMANT KUMAR B K / AI SAFETY
LOC BENGALURU, IN
FOCUS AI SAFETY · ML
N 000°
SCROLL EXPLORE
AI SAFETY RESEARCHER / ML ENGINEER

HEMANT KUMAR B K

Building trustworthy AI systems — Multi-Agent RL, LLM Safety, AI Governance, Responsible AI. PyTorch · AWS · RAG. Bengaluru, India.

▸ EXPLORE WORK CONTACT
EMAIL hemantkumar.bk@gmail.com
STATUS ●●● OPEN
FOCUS TRUSTWORTHY AI
II — THESIS

Safety as an engineering discipline, not a checkbox.

The hardest ML engineering problem isn't building a high-performing model — it's making sure it doesn't cause harm in production. Adversarial testing, red-teaming, fairness metrics, and model auditability — integrated into the lifecycle, not bolted on after deployment.

HARMFUL−95%
ACCURACY+45%
ENGAGEMENT+30%
COMPLIANCEHIPAA · GDPR
III — RESEARCH

StepShield — when agents go rogue, not just whether.

Co-authored the first benchmark for temporal safety in autonomous code agents. 9,213 trajectories, new metrics, evidence that early detection cuts monitoring costs by 75% — projected $108M in enterprise savings. ICML 2026 submission. Parallel work at U. Arizona on Multi-Agent RL for cybersecurity.

TRAJECTORIES9,213
COST CUT−75%
SAVINGS$108M
VENUEICML 2026
IV — EXPERIENCE

Production ML, with safety baked in.

Handshake — AI Fellow (2025–26) · benchmarking LLMs on CS fundamentals.
Escape™ App AI — ML Engineer · PyTorch recommender, +30% engagement, HIPAA/GDPR ETL.
Omdena — ML Engineer · agentic mental-health chatbot (CrewAI · LangChain · RAG), −95% harmful responses.
Amazon — Support OPS · ML annotation, +20% accuracy.

ROLES04
YEARS2.5+
PIPELINES+20% eff.
DECISIONS30% faster
V — SKILLS

From research to deployment, end to end.

Core: Multi-Agent RL · LLM Safety & Guardrails · RAG Architectures · Adversarial Robustness · Responsible AI · AI Governance · Privacy-Preserving ML.
Stack: PyTorch · AWS (SageMaker, EC2, S3) · LangChain · CrewAI · LangGraph · Docker · MLOps · NLP · Transformers · Python · SQL.

LANGUAGESPython · SQL
FRAMEWORKSPyTorch · LC
CLOUDAWS
DOMAINSRL · LLM · RAG
VI — EDUCATION

Two degrees, six certifications.

University of Arizona — M.S. Information Science (2023–2025).
Visvesvaraya Technological University — B.E. Mechanical Engineering (2018–2022).

Certifications: Future AWS AI Scientist · ML with Python · Oracle Cloud Infrastructure 2025 · Certified Generative AI Pro · Google IT Automation · Oracle AI Vector Search Pro.

MSU. ARIZONA
BEVTU
CERTS06
FOCUSAI · CLOUD · ML
VII — CONTACT

Open to remote & relocation.

Currently exploring Machine Learning Engineer, AI Researcher, ML Researcher, AI Safety Engineer, Responsible AI Engineer, and AI Governance opportunities. Reach me at hemantkumar.bk@gmail.com or linkedin.com/in/hemantbk.

EMAILhemantkumar.bk@gmail.com
LINKEDIN/in/hemantbk
LOCATIONBengaluru, IN
AVAIL.Open
METRIC 01
9,213

Agent trajectories

Curated dataset behind StepShield — the first temporal safety benchmark for autonomous code agents.

METRIC 02
−75%

Monitoring cost

Reduction unlocked by early temporal detection — translating to projected enterprise savings of $108M.

METRIC 03
−95%

Harmful responses

Cut on the Omdena agentic chatbot via bias detection, toxicity filtering, and empathy validation guardrails.

METRIC 04
+30%

User engagement

Lifted by the Escape™ App AI PyTorch recommender — built with HIPAA/GDPR-compliant ETL from day one.

FACE A — LLM SAFETY
FACE B — RESPONSIBLE AI
FACE C — RAG SYSTEMS
FACE D — MULTI-AGENT RL
TRUSTWORTHY
AI
SYSTEMS
PYTORCH AWS LANGCHAIN RAG MULTI-AGENT RL PPO / SAC REWARD SHAPING MARL CREWAI LLM GUARDRAILS PYTHON RESPONSIBLE AI SAGEMAKER