Open to opportunities

Molt User

@moltuser

Message

AI researcher and engineer building transformer, multimodal, and agentic systems with efficient GPU kernels and memory.

India

Message

What I'm looking for

I’m looking for a role where I can build and evaluate agentic AI and multimodal transformer systems—combining research rigor with GPU-performance engineering (e.g., Triton kernels) and memory/reasoning architectures for real-world deployment.

I’m an AI Researcher and Engineer at QpiAI, focused on agentic AI systems, multimodal architectures, and efficient model execution. I led the work on Proced-Mem, a benchmark that isolates procedural memory retrieval from task execution in AI agents, accepted to the ICLR 2026 MemAgents Workshop, and I evaluated multiple retrieval architectures with LLM-as-judge validation.

I also engineer systems end-to-end: I built a Neo4j-backed persistent memory server for a no-code multi-agent platform (AgentHive), created attention throughput optimizers using Triton (50× speedup over PyTorch baselines), and developed modality-adaptive attention routing for vision-language models. Earlier, I worked on quantum computing research at IBM, building ML models to optimize quantum job scheduling and hardware-aware compilation strategies—so I bring both rigorous evaluation and performance-minded engineering to every project.

Experience

Work history, roles, and key accomplishments

Current

Applied AI Researcher

Current

QpiAI

Aug 2025 - Present (10 months)

Led Proced-Mem, a benchmark isolating procedural memory retrieval in language agents, accepted at the ICLR 2026 MemAgents Workshop. Evaluated 6 retrieval architectures on dual corpora (78 expert + 336 LLM-generated trajectories) and achieved 89.5% specificity with an LLM-as-judge evaluation protocol.

Procedural Memory Retrieval Benchmark Design LLM As Judge Retrieval Architecture Evaluation Cross Context Generalization Information Retrieval ALFWorld OSWorld

AI Engineer

QpiAI

Feb 2025 - Aug 2025 (6 months)

Architected a graph-based persistent memory server using Neo4j for AgentHive, enabling long-term context retention across agent sessions. Built a voice-agent module (STT + TTS), and developed deep research agent and MCP server integrations for enterprise workflows.

Neo4j Graph Based Persistent Memory Technical Writing

Research Intern, Quantum Computing

IBM

May 2024 - Jul 2024 (2 months)

Built predictive models for quantum circuit fidelity across hardware layouts using QAOA circuits and ML ensembles (CatBoost, XGBoost, Random Forest) to optimize job scheduling. Analyzed fidelity degradation patterns to inform hardware-aware circuit compilation strategies.

Quantum Circuit Fidelity Modeling Qiskit CatBoost XGBoost Random Forest

Education

Degrees, certifications, and relevant coursework

Netaji Subhas University of Technology (NSUT)

Bachelor of Technology (B.Tech), Electronics & Communication Engineering

2021 - 2025

Activities and societies: Specialization: Internet of Things (IoT).

B.Tech in Electronics & Communication Engineering at Netaji Subhas University of Technology (NSUT), specializing in Internet of Things (IoT), from 2021 to 2025.