Pragnyan Ramtha

about me.

Results-driven AI/ML Engineer specializing in Large Language Model (LLM) fine-tuning and AI system design. I design maintainable, production‑grade AI systems and can comfortably work with deep cloud infrastructure. I learn new tools fast and use AI as a force‑multiplier in my coding, designing, and research loops, which lets me move much faster while keeping systems reliable.

experience.

• AI Engineering Intern Remote

  at, Reputation-DAO

  Aug 2025 – Jan 2026

  • Architected a high-availability serverless MLOps backend on GCP (Cloud Functions, Cloud Run) for production grade AI orchestration, achieving 99.9% uptime while successfully cutting down on cloud costs by 70%.
  • Engineered a Gemini API response system with optimized prompt caching, which significantly reduced API costs and delivered a 50% reduction in inference latency across critical workflows.
  • Boosted accuracy of customer support bot by developing a RAG pipeline utilizing semantic search for real-time documentation retrieval and source attribution.

  • GCP
  • Gemini API
  • RAG
  • Serverless
  • Python

• Github Contributor Remote

  at, Open Source Contributions

  Jan 2025 – Present

  • Served as a key contributor to 30+ open-source projects, focusing on bug fixes and core feature development.
  • Identified a bottleneck in Scrapy causing severe latency, authored a fix that eliminated the drag and achieved a 2x speedup for affected workflows.
  • Winner of IEEE Summer of Code (IEEESoC) Hackathon 2025, for open source contributions to multiple projects.

  • Python
  • TypeScript
  • Git
  • Docker
  • CI/CD

projects.

• AIMO-3: Efficient Reasoning via LLM Fine-Tuning

  View Project

  • Fine-tuned Phi-4 (14B) on CoT and TiR datasets to optimize multi-step problem solving and tool-use efficiency.
  • Achieved 90% accuracy on reasoning benchmarks, rivaling 125B parameter models while utilizing significantly fewer compute resources.

  • Phi-4
  • Fine-tuning
  • CoT/TiR
  • PEFT

• Personality Clone

  View Project

  • Fine-tuned a Large Language model, leveraging PEFT (QLoRA) and contrastive learning on private conversational data to emulate personal response style.
  • Implemented a siamese network architecture with cosine similarity loss, which improved semantic embeddings and achieved 92% accuracy in replicating my response style, a 28% improvement over baseline models.

  • TensorFlow
  • Python
  • CUDA
  • Transformers

• Autopilot

  View Project

  • Engineered an AI-driven OS automation system leveraging function calling and tool-use paradigms to execute complex natural language tasks to achieve low-level automation.
  • Built a Reasoning + Acting agent framework with command sandboxing, reducing execution errors and achieving 45% faster task completion than manual workflows.

  • Python
  • LLM Agents
  • APIs

blogs.

• Why OpenAI Sent Me $500 for a Research ProjectApr 2026

  Read full post 8 min read
  • A complete breakdown of how I built XSA4 + EMA + GPTQ-Int6, the submission that landed at 1.1271 BPB and placed top-5 globally in OpenAI's Parameter Golf challenge.
  • Every technique explained from first principles: what bits-per-byte is, how Cross-Sparse Attention works, why EMA produces better weights than a single checkpoint, and what GPTQ actually does differently from naive quantization.
  • Parameter Golf
  • OpenAI
  • Cross-Sparse Attention
  • GPTQ
  • Model Compression

• How I Reached #1 on ARC-AGI-2Apr 2026

  Read full post 7 min read
  • How I adapted the parallel agent and budget allocation patterns I built for AIMO3 to reach the top of the ARC Prize 2026 leaderboard, using per-puzzle test-time training, a vocabulary-restricted DFS beam search with KV cache reuse, and augmented re-scoring.
  • A breakdown of the full pipeline: what I tried, what finally worked, and the three engineering tricks that made the difference.
  • ARC-AGI-2
  • Test-Time Training
  • Qwen
  • Kaggle
  • Reasoning

• How I Won a Solver Medal at AIMO3Apr 2026

  Read full post 6 min read
  • A walkthrough of the agentic system I built for AIMO3, the $2.2M Kaggle competition to make AI solve International Mathematical Olympiad problems, using Gemma-4-31B, parallel sandboxed code execution, weighted voting, and a budget-aware time allocator.
  • What I tried, what flopped, and the three engineering tricks that actually moved the needle.
  • AIMO3
  • Gemma 4
  • Math Reasoning
  • Agentic LLMs
  • Kaggle

research papers.

• Scaling Context Windows to Infinity: A Comprehensive Study of Position Encoding, Attention Mechanisms, Memory-Efficient Inference, and Context Reduction Techniques in Large Language Models 2026

  Read paper Academia.edu

  A comprehensive analysis of techniques for extending context windows in large language models, examining position encoding strategies, efficient attention mechanisms, and memory-optimized inference approaches to enable processing of arbitrarily long sequences.

• Unlocking Societal Trends in Aadhaar Enrolment and Updates: Anomaly Detection and Fraud Risk Prediction 2026

  Read paper Academia.edu

  A data-driven approach to identify suspicious patterns in India's Aadhaar biometric identification system, utilizing machine learning for anomaly detection and fraud risk prediction in enrollment and update processes.

• Speeding Up LLM Inference Using Quantum Computing Techniques 2026

  Read paper Under Research

  Exploring quantum-inspired algorithms and quantum computational primitives to accelerate inference in large language models, investigating quantum annealing for attention mechanisms and variational quantum circuits for efficient token generation.

technical skills.

• Languages:

  Python, C++, Bash, SQL, TypeScript

• AI/ML Tools:

  PyTorch, Transformers, Unsloth, NumPy, Pandas, Scikit-learn, PEFT/QLoRA, CUDA

• Infrastructure:

  GCP, Azure, Docker, Linux (Arch), Git

• Certifications:

  Machine learning certification (Stanford), CS50: comp. Sci. (Harvard University)

• Achievements:

  • Artificial Intelligence Mathematical Olympiad (AIMO) Silver Medalist
  • Authored 2 Research Papers on Modern AI Optimization Techniques
  • Winner, IEEE Summer of Code (IEEESOC) Hackathon 2025
  • Winner, Empathy Encryption Hackathon 2025
  • Winner, Daydream Hyderabad @ Hackclub 2025
  • Top 0.5% Finalist, Shell AI Hackathon 2025

• Developer Tools:

  uv, Neovim, Arch Linux