Open to residencies & fellowships
Pranav Dhiran
›|
Electronics engineer turned AI builder. I work across the stack — from transformer pre-training and RL fine-tuning to shipping multi-agent systems.
Committed to building AI solutions that solve real-world challenges.
01
Education
2023 — Present
Shri Guru Gobind Singhji Institute of Engineering & Technology, Nanded
B.Tech — Electronics & Telecommunication Engineering
Minor in Information Technology
02
Selected Work
Small Language Model From Scratch — TinyStories
Pre-trained a GPT-style autoregressive transformer (6L, 6H, 384-dim) on TinyStories from scratch — custom BPE tokenization, mmap pipelines, AMP mixed precision, gradient accumulation, warmup + cosine LR scheduling, temperature/top-k sampling.
PyTorchHugging FaceAMPNLP
→
Medaura — Agentic Pharmacy System
Full-stack multi-agent AI system — five specialized agents (Ordering, Safety, Forecast, Procurement, UI) with a custom orchestration pipeline for autonomous medication ordering across four languages.
FastAPILangChainChromaDBLangfuseReact
→
GNU Radio MCP Server — LLM-to-SDR Bridge
MCP server bridging LLMs to live GNU Radio SDR flowgraphs — 13 tools with Pydantic v2 validation, lifespan-managed ZMQ context, and stdio/streamable-HTTP transport. Includes async IQ capture with Welch PSD, peak detection, and automated frequency sweep.
PythonFastMCPZMQXML-RPCGNU Radio
→
RF Watch — Open-Source Real-Time RF Spectrum Monitor
Real-time RF spectrum analyzer using HackRF One + GNU Radio — FFT-based spectral feature extraction with a lightweight ML classifier for passive detection of unknown transmitters. Built on work from SIH 2025, originally developed as an anti-drone detection system for ITBP.
PythonGNU RadioHackRF OneSignal Processing
→
03
Technical Skills
Languages & Frameworks
PythonPyTorchTensorFlowHugging Face Transformers
LLM Engineering
Instruction Fine-TuningLoRAPEFTINT4/INT8 QuantizationUnslothLangChainLangGraphOpenAI/Gemini APIsChromaDBFAISS
Agentic & MCP
Multi-Agent SystemsTool-UseFunction CallingMCP Server EngineeringFastMCPXML-RPCZMQLangSmithLangfuse
04
Research & Achievements
Research Interests
RLHFRLAIFGRPOScaling LawsMoESSMsEfficient InferenceMulti-Agent ReasoningTokenization
Certifications
• Fine-tuning & RL for LLMs: Intro to Post-training — DeepLearning.AI
• AI Agents in LangGraph — DeepLearning.AI
• Quantization Fundamentals — Hugging Face
• MCP: Build Rich-Context AI Apps — Anthropic
Awards & Recognition
• Global Finalist (Top 6 Internationally) — UWA Hack For Impact 2026
• National Finalist — Smart India Hackathon (SIH) 2024 & 2025
• Regional Qualifier — Nxt Wave x OpenAI Buildathon
05