Shivom Hatalkar
Building Intelligent Systems at the Intersection of AI, Language & Human Experience
Who I Am
About Me
LLM / GenAI Engineer specializing in fine-tuning and deploying large language models, and building agentic AI systems for real-world, production SaaS environments. Experienced in designing multi-tool agent architectures, RAG pipelines, and full-stack AI products. Hands-on expertise in LoRA/QLoRA-based training, LLM tool calling/orchestration, Hugging Face, and cloud-based GPU infrastructure. Passionate about developing scalable, business-impactful AI solutions in domains like HR automation, healthcare, and finance.
3B+
LLM Params Fine-tuned
23.7K
Training Dialogues
5+
AI Projects Shipped
LLM Fine-Tuning
Hands-on with LoRA/QLoRA on 3B+ parameter models
RAG Systems
Production-ready retrieval-augmented generation pipelines
Full-Stack AI
End-to-end AI applications with Next.js & Python
Agentic AI
Multi-agent workflows with N8N and LLM orchestration
Academic Background
Education
University of Mumbai
Bachelors of Engineering
Artificial Intelligence & Machine Learning
- B.E. in Artificial Intelligence & Machine Learning, and Honors in Cyber Security
- Specialized in AI/ML with a focus on deep learning, NLP, and agentic systems
- Built multiple production-ready AI pipelines and fine-tuned LLM architectures
Thakur Polytechnic
Diploma in Computer Engineering
Computer Engineering
- Foundation in computer science, software engineering, and database systems
- Developed core programming and systems engineering knowledge
B.E. in AI & ML — University of Mumbai
Specialised in Artificial Intelligence, Machine Learning, and Cyber Security — building the theoretical and practical foundation for a GenAI career.
Work History
Experience
Agentic AI Engineer Intern
Buzzworks Business Services Pvt. Ltd.
- 1Building Buzzbot, an AI chatbot assistant for Veytan (Buzzworks' HRMS/HCM SaaS platform) that resolves policy queries, guides users through the platform, and executes actions such as applying for leave and regularizing attendance.
- 2Designed the agent's backend orchestration layer around DeepSeek V4 Flash as the reasoning 'brain', wiring it to policy-document, attendance, employee-details, and leave-application APIs so it autonomously selects and calls the right tool per query.
- 3Owned the full tool-calling loop and agent state entirely server-side, exposing a single Buzzbot API that the frontend widget calls — keeping the client thin and the reasoning/orchestration logic centralized and maintainable.
- 4Developing an Exit Agent for Veytan's Exit module that ingests and parses incoming exit-related emails (including OCR on attachments), auto-fills the exit-initiation form for Ops/Manager approval, then tracks and manages downstream exit tasks and generates exit letters — an end-to-end agentic workflow for employee offboarding.
Machine Learning Intern
Prodigy Infotech
- 1Developed and evaluated ML models including Linear Regression, SVM, and K-Means on structured datasets.
- 2Performed data preprocessing, feature engineering, and model evaluation to improve prediction performance.
Web Development Intern
Nibodh Educare
- 1Developed and maintained responsive web application components across frontend and backend.
- 2Worked with HTML, Tailwind CSS, React.js, Python, Django, and MySQL for full-stack development.
What I've Built
Projects
From fine-tuned 3B-parameter mental health LLMs to AI-powered finance and fitness platforms — click any card for architecture details, engineering challenges, and key learnings.
RAGiment
Open-Source RAG Pipeline Generator
A wizard-driven tool that turns a short intake (corpus size, latency, cost, and infra constraints) into a production-ready RAG pipeline — runnable code, vector-DB setup, config, and an evaluation suite.
6
RAG Architectures
4
Frameworks
5
Vector DBs
2 (Static/BYOK)
Codegen Modes
Elixir-v2
3B-Parameter Mental-Health Conversational Model
A domain-specialized conversational AI model fine-tuned from Llama-3.2-3B-Instruct, designed specifically for mental health support conversations. Built with enterprise-grade model governance and responsible AI principles.
23.7K
Training Dialogues
3B
Model Parameters
RTX 3090
GPU
6+
Eval Metrics
Tree of Truth
RAG-Powered Guide to Consciousness & Reality
An interactive RAG-powered application bridging Advaita Vedanta, philosophy of mind, and cognitive science using retrieval-augmented generation over non-dual literature.
Full-Stack
Stack
Vector DB RAG
Search Engine
Next.js
Framework
Consciousness AI
Domain
Finance Tracker
AI-Enhanced Budgeting Tool
A full-stack personal finance management application with AI-powered spending insights. Integrates Mistral-7B to generate intelligent suggestions for optimizing spending habits in real-time.
Full-Stack
Stack
Mistral-7B
AI Model
Clerk
Auth
Drizzle
ORM
Pulse Fitness
Intelligent Fitness & Nutrition Assistant
An AI-powered fitness and nutrition assistant that generates personalized exercise and diet plans using Gemini LLM. Combines voice-driven AI interaction via Vapi with a polished Next.js frontend.
Gemini
AI
Vapi
Voice
Convex
Backend
AI SaaS
Type
Technical Expertise
Skills
AI / ML & GenAI
Data & Analytics
Frameworks & Tools
Cloud & Platforms
Web & Backend
Domain Proficiency Overview
Tools & Technologies
Credentials
Certifications
Click any completed certificate to view the PDF
IBM
IBM RAG and Agentic AI Professional Certification
Google Cloud
Google Cloud Platform (Vertex AI, Gemini)
4
Completed
2
In Progress
6
Total
Thoughts & Musings
Blog
Exploring the Boundaries of Mind & Machine
Beyond code, I explore Advaita Vedanta, consciousness studies, cognitive science, philosophy of mind, and AI & consciousness.
The Hard Problem of Consciousness: Where AI Meets Advaita
Exploring how Chalmers' hard problem of consciousness resonates with the Advaita Vedantic concept of Brahman as pure awareness — and what this means for AI systems that simulate understanding.
Large Language Models: Syntax Without Semantics?
Do LLMs genuinely understand language, or are they sophisticated pattern matchers? A cognitive science lens on the distinction between syntactic processing and semantic grounding.
Non-Dual Awareness and the Observer Effect in Quantum Cognition
Drawing parallels between the Advaitic notion of the witness-consciousness and quantum observer effects — how observation shapes reality in both physics and mind.
Building Empathetic AI: Lessons from Mental Health Conversational Models
Reflections from building Elixir-v2 — what it taught me about empathy, ethics, and the responsibility of deploying AI in sensitive domains.
The Great Retrieval Debate: Vector vs. Vectorless RAG
At the dawn of new emerging technologies, which type of RAG will gain the most trust from the enterprises - will cosine similarity win or will semantic tree graphs take the lead?
Get In Touch
Contact
Open to GenAI, Data & AI, and full-stack AI engineering opportunities. Feel free to reach out!
Seeking GenAI / Data & AI roles. Response time: within 24 hours.