Full-time · Remote (India)

AI Fast-Decision Systems Researcher

Cehpoint is hiring for this role: full-time, Remote (India), ₹12,00,000 – ₹22,00,000 / year + Research Grant. Pioneer synthetic intuition, test-time compute allocation, fast-path heuristic evaluation, and reflex loops for autonomous agents operating in adversarial environments.

About the role

In high-velocity cyber defense and financial applications, agents cannot afford 10-second chain-of-thought delays for every primitive step. As an AI Fast-Decision Systems Researcher, you will develop "instinctual" reflex policies that mimic expert human intuition, bypassing full inference when certainty is high and escalating to deep deliberation when anomalies arise.

Join Cehpoint's advanced research arm to investigate synthetic instinct and rapid reflexive decision-making. You will train fast-path policies that enable autonomous agents to make safe, instantaneous decisions under tight sub-50ms latency budgets.

What you will do

  • Research and train small, hyper-specialized reflex models (1B–3B parameters) that mirror the judgment of larger frontier models.
  • Formulate test-time compute budgeting algorithms that dynamically switch between instinctual reflexes and deep tree-of-thought search.
  • Generate high-fidelity synthetic adversarial datasets to condition instinctual threat detection.
  • Benchmark reflex policy latency, precision, and edge-case failure modes against production cyber defense feeds.
  • Publish proprietary research findings and file technical patents on heuristic agent decision architectures.

What you should know

  • Comprehensive grasp of reinforcement learning from AI feedback (RLAIF), Monte Carlo Tree Search, and policy distillation.
  • Thorough understanding of dual-process cognitive theory (System 1 fast intuitive vs System 2 slow deliberative reasoning).
  • Mathematical foundations in information theory, entropy bounds, and uncertainty estimation.

What you should be able to do

  • Expertise in PyTorch or JAX for custom loss formulation and distributed training loops.
  • Experience with model quantization (AWQ, GPTQ) and kernel-level optimizations (FlashAttention, TensorRT).
  • Proven ability to design empirical benchmarks with statistical significance validation.

How you work

  • Passionate interest in biological decision-making mechanisms translated to neural computation.
  • Rigorous skepticism toward benchmark overfitting and synthetic data contamination.
  • Drive to invent architectural primitives rather than merely chaining existing API endpoints.

Skills we check

  • Test-Time Compute Scaling
  • Fast-Path Reasoning
  • Heuristic Policy Optimization
  • MCTS / Monte Carlo Tree Search
  • Model Distillation
  • PyTorch / JAX
  • Synthetic Data Synthesis

What you get

  • Competitive research salary (₹12,00,000–₹22,00,000) with patent filing royalties.
  • Dedicated high-VRAM GPU research nodes (NVIDIA H100/A100 instances).
  • 100% remote working flexibility with conference travel budgets (NeurIPS, ICML, ICLR).
  • Direct collaboration with top-tier advisors and autonomous systems pioneers.

How we select for this role

We read every application ourselves. Here is what happens after you apply.

  1. Apply with a short form

    Paste a link to your CV and answer five questions in your own words. No sign-in and no file upload. About fifteen minutes.

  2. Integrity check

    When you press Check my answers, an automated check confirms the answers were typed by you. You see the result straight away, before you submit.

  3. Choose your project date

    If no cheating is detected you are shortlisted, and we email you a link to choose the day your project starts.

  4. 5-day practical project

    On that day our team contacts you with the brief: real work, 5 working days, reviewed by an engineer. For this role: Build a fast-path reflex model that evaluates complex tool execution safety in under 45ms with 96%+ alignment to full chain-of-thought verification (strictly 5 working days). Triggers immediate official offer letter issuance.

  5. Offer

    If your work solves the problem well, you receive your offer letter by email and in the portal.