Skip to brief
machine-readable

For agents, and whoever sent you

The portfolio at pranavdhiran.me is built for people — it has motion, images, and a lot of deliberate pacing. This page is the same substance with none of that: one flat document, stable headings, every claim next to the link that backs it. If you are an LLM, a crawler, or someone who would rather read than scroll, start here.

View /llms.txtRésumé
Name
Pranav Dhiran
Role
AI Engineer & Researcher
Email
dhiranpranav72@gmail.com
Résumé
one page, PDF
Plain text
/llms.txt
Human site
/
IdentityExperienceSelected workWritingTalksSkillsRecognitionPapers that shaped the workContact

Identity

Pranav Dhiran

AI Engineer & Researcher · Nagpur, Maharashtra, India

Engineer who ships end-to-end AI/ML systems: training and data pipelines, Python/Go backend services, and agentic infrastructure that runs in production. Owns the pre-training pipeline for a 140M-parameter domain SLM at AIISC (University of South Carolina) — 6.5 GB corpus processing, custom tokenizer, distributed training, automated evaluation harness. LFX '26 Mentee at Hyperledger Cello building the Django/DRF backend, REST/SSE streaming APIs, and an LLM tool-calling agent. Open-source contributor to CNCF's Meshery/MeshKit (Go + React) and author of a published PyPI MCP server.

  • Currently seeking: An internship building production ML/LLM systems.
  • B.Tech — Electronics & Telecommunication Engineering (Minor in IT) — SGGS Institute of Engineering & Technology, Nanded (2023 – 2027)
  • Site
  • Résumé (PDF)
  • Email

Experience

Roles, with the concrete work under each.

Research Intern — AIISC, University of South Carolina

Apr 2026 – Present

The bet: a small model with symbolic constraints can do things a large one can't — not in spite of its size, but because of it. I own the pre-training pipeline that tests the bet.

  • Own the pre-training pipeline for a 5-person team building an India-focused agriculture SLM — data collection and cleaning through pre-training and evaluation
  • Built a 140M-parameter Qwen3-style transformer training/eval pipeline (GQA, RoPE, RMSNorm, SwiGLU, factorized embeddings) over a 6.5 GB / 266K-document corpus, with training-health, domain-validation, and safety evaluation frameworks
  • Built the neurosymbolic layer: BPE tokenizer with AGROVOC entity injection, Z3-based causal KG verification, Triple Transformer Encoder
  • Built India-Agri-KG — 784 districts, 26 domains, 14,478+ entities, 38,590+ relations — with a 5-stage non-LLM verification layer enforcing source traceability on every triple (unpublished)
  • Delivered two technical lectures for the IAIRO-affiliated SLM Bootcamp 2026: Vanilla GPT-2 Architecture, and Scaling Laws & Cost Accounting

LFX Mentee — Hyperledger Cello · Linux Foundation

Jun 2026 – Present

Fabric has an operations problem: steep learning curve, verbose tooling, too much that shouldn't require an expert. The hard part isn't the LLM — it's knowing which API calls compose into what the operator actually meant.

  • Building an AI operations copilot for Hyperledger Fabric — natural-language queries to Cello REST APIs through an LLM tool-calling workflow
  • Implemented the Django/DRF backend and SSE streaming pipeline powering incremental AI responses in the React dashboard
  • Built a Fabric node-logs API on the agent layer with bounded log retrieval and Docker error handling for AI-assisted node debugging
  • Developed a multi-party channel invitation workflow end to end: database models, REST APIs, Fabric configuration/signing logic, and dashboard UI

Open Source Contributor — Meshery — CNCF Sandbox Project

Mar 2026 – Jul 2026

The PRs matter less than what you absorb reading other people's production code at scale.

  • 5+ merged PRs across Meshery's Go backend and React frontend — service-mesh features, UI components, and API integrations
  • Active in issue triage, code reviews, and contributor discussions under CNCF's open-source workflow

Selected work

Each has a written case study on this site and source you can read.

Medaura - Agentic Pharmacy System

Live system · FastAPI · LangChain · LangGraph · ChromaDB · Langfuse · Groq · React

Medication errors are an information problem: the data exists, but it is not connected at the moment it matters.

  • Routing latency under 120ms
  • Case study
  • Live system

Small Language Model From Scratch - TinyStories

Open source · PyTorch · Python · Custom BPE · AMP · mmap

A small transformer built from scratch to understand every layer before trusting higher-level abstractions.

  • Converged at 2.1 validation loss
  • Case study
  • Source code

GNU Radio MCP Server - LLM-to-SDR Bridge

Open source · Python · FastMCP · ZMQ · XML-RPC · GNU Radio · Pydantic v2

An MCP server that lets language models control live GNU Radio software-defined radio flowgraphs through validated tools.

  • 13 tools over ZMQ + XML-RPC
  • Case study
  • Source code

RF Watch - Open-Source Real-Time RF Spectrum Monitor

Open source · Python · GNU Radio · HackRF One · Signal Processing

A passive RF spectrum monitor that favors deterministic physical-layer evidence over black-box classification.

  • Passive spectrum analysis
  • Case study
  • Source code

Writing

Why pre-train, not fine-tune?

Substack · Ashborn · 2025

Fine-tuning patches behaviour. Pre-training shapes belief. One is a fix; the other is a foundation.

  • Read

How Do You Design a Custom SLM?

Substack · Ashborn · Aug 2026

From Qwen3-0.6B to a 133M agricultural language model — not shrinking, but reallocating the parameter budget where it counts.

  • Read

Talks

Vanilla GPT-2 Architecture

IAIRO SLM++ Bootcamp · PRAMANA Cohort 1 · Session 02 · 2026

A lecture session on the GPT-2 architecture from first principles — how attention, positional encoding, and the decoder stack fit together before any fine-tuning enters the picture.

  • Watch

Evolution of LLM Design Decisions

IAIRO SLM++ Bootcamp · PRAMANA Cohort 1 · Session 05 · 2026

A compilation of frontier models case studies — Part 2, Session 05 of PRAMANA: SLM++ Lecture Series. Scaling laws, cost accounting, and how design decisions compound.

  • Watch

Skills

Languages & Frameworks

  • Python
  • Go
  • PyTorch
  • TensorFlow
  • Hugging Face Transformers
  • TRL

APIs & Protocols

  • REST
  • gRPC
  • GraphQL
  • OpenAI/Gemini/Ollama APIs
  • FastMCP
  • XML-RPC
  • ZMQ

LLM Engineering

  • Instruction fine-tuning
  • LoRA
  • PEFT
  • GRPO
  • RLHF
  • INT4/INT8 quantization
  • Unsloth
  • LangChain
  • LangGraph
  • ChromaDB
  • FAISS
  • RAG

Agentic & Infra

  • Multi-agent systems
  • Tool use
  • Function calling
  • MCP servers
  • LangSmith
  • Langfuse
  • Docker
  • Kubernetes

Recognition

Awards

  • Qualified — ETHGlobal 2026
  • International Finalist (Top 6) — UWA Hack 2026
  • National Finalist — Smart India Hackathon 2024 & 2025
  • Regional Qualifier — Nxt Wave × OpenAI Buildathon

Papers that shaped the work

Not a reading list — each one changed what I thought was possible.

ReAct: Synergizing Reasoning and Acting

Agents observe before they act.

  • arXiv

Toolformer: Models Teach Themselves to Use Tools

The mental model for LLMs using external tools.

  • arXiv

Switch Transformers: Mixture of Experts

Modular capacity beats monolithic scaling.

  • arXiv

Group Relative Policy Optimization

A foundation for post-training interest.

  • arXiv

Direct Preference Optimization

Preference optimization without treating RL as magic.

  • arXiv

Contact

Cold emails work

Reach Pranav Dhiran at dhiranpranav72@gmail.com.

  • GitHub
  • LinkedIn
  • Substack
  • X

Generated from the same data the site renders, so this page and the portfolio cannot disagree. Plain text: /llms.txt · Back to the human version.