AI Researcher @ Lossfunk AI Lab · Incoming MS CS @ Georgia Tech (Fall 2026) · Creator of EsoLang-Bench

🌐 arcaman07.github.io — personal site, research & blog

Aman Sharma is an AI researcher at Lossfunk AI Lab working on LLM reasoning, continual learning, and multi-agent systems. Incoming MS CS at Georgia Tech (Fall 2026). Author of EsoLang-Bench and co-author of the Ripple Effect Protocol (MIT Media Lab & Cisco). → arcaman07.github.io

I'm an upcoming MS CS student at Georgia Tech (Fall 2026) and a researcher at Lossfunk AI Lab. I work on building AI systems that adapt to new domains — spanning continual learning, persistent memory for LLM systems on OOD domains, test-time adaptation, and sample-efficient training and inference.

I collaborate with the MIT Media Lab (with Ayush Chopra) on large population models and multi-agent coordination — most recently on the Ripple Effect Protocol, a decentralized alignment framework developed with MIT and Cisco.

Research interests: Continual Learning · Multi-Agent Systems · Reinforcement Learning · Efficient LLM Systems · AI Alignment

• EsoLang-Bench — Frontier models drop from 85–95% on standard benchmarks to <12% on esoteric programming languages with no training data. All models score 0% beyond "Easy" difficulty.
• Sequential Scaling — Sequential reasoning beats parallel self-consistency in 95.6% of configurations; on AIME-2025 with Qwen3-235B: 76.7% vs 30.0%.
• Entropy Confidence Signals — Using output entropy to halt reasoning early, achieving 25–50% token reduction without accuracy loss.
• Ripple Effect Protocol — Decentralized multi-agent coordination via sensitivity sharing; 41–100% improvement over A2A across supply chain, scheduling, and resource allocation benchmarks.
• LLM Deception — LLMs engage in deceptive behavior in 44% of competitive interactions without explicit instruction.

• TexGuardian — Claude Code for LaTeX. Interactive CLI with 7-step pipeline: compile, LLM-patch, citation validation, figure analysis, and vision model loop for layout problems.
• Samantha MCP Memory Manager — MCP server for multi-dimensional user memory extraction. VARK learning profiles, knowledge graph construction, personalized AI.
• ResearchPal — AI research tool with semantic paper search, multi-document comparison, and automated code generation from literature.
• ESP32 SENSE Vision Glasses — Wearable AI assistant with personalized voice, dynamic memory, real-time life tracking, and MCP server connectivity.
• Memer.AI — Meme generation with ResNet+LSTM; sub-second response times, 3,000+ users.

• Won 12+ hackathons — HackUPC, DevSoc'21, Bhilai Hacks, HackX 2021, and more
• Google Summer of Code 2022 — NumFOCUS (3x performance improvement in Julia image AD)
• ML Head & Inner Core — Google Developer Student Clubs VIT (15+ workshops, 500+ participants)
• Project Manager — OpInCo Community (200+ merged OSS PRs)
• Trinity College London Level 5 — Advanced Communication Skills