19 Nov
|
Disseqt AI
|
Bengaluru
19 Nov
Disseqt AI
Bengaluru
The Role You will build the science behind our platform
the attacks that break AI systems, and the models that stop them. This is genuine research with a short path to production. Your work ships into live enterprise deployments, gets scrutinised by auditors and regulators, and can be published. What you'll work on
• Adversarial research
designing jailbreak agents, multi-turn attacks, and measuring how attacks transfer across models
• Validator models
training, fine-tuning, and calibrating the models that catch unsafe AI behaviour in real time
• Evaluation methodology
building the statistics behind claims like “this model is secure”
• Interpretability & fairness
explaining why a model failed, in a form an auditor accepts
• Recent attack surfaces
multi-modal, RAG, and agentic tool-calling, where almost no methodology exists yet What We're Looking For
• Transformers
you understand attention, tokenisation, and model internals,
not just the API
• Classical ML
supervised learning, ensembles, calibration, precision/recall trade-offs, imbalanced data. This matters as much as the LLM work.
• Fine-tuning
SFT, LoRA/PEFT, distillation. Awareness of RLHF, reward modelling, and DPO.
• Statistics
experimental design, significance, confidence intervals
• Python
clean, reproducible code. PyTorch, HuggingFace, scikit-learn. Who You Are
• Curiosity about how systems break
• Comfort with problems that have no textbook answers
• Transparent writing
you can explain your work to a researcher and to a compliance officer Bonus points
• Publications or preprints
• Red-teaming or CTF experience
• Open-source contributions
• Security background
📌 Sr Ai Research Scientist Bengaluru
🏢 Disseqt AI
📍 Bengaluru