Om Patil
Building AI Systems That Go Beyond the Demo.
Introduction
Building AI Systems That Go Beyond the Demo.
AI & Data Science engineer focused on Generative AI, Agentic Systems, and full-stack AI products — building multi-agent research workflows, RAG systems, and LLM products that ship.
Social links
Overview
Director, Product, Strategy & Technology @AskLumenAI
Student Founder — Product & Full-Stack Engineer @ViksitHub
GitHub contributions
About
- I'm Om Patil — an AI & Data Science engineer focused on Generative AI, Agentic Systems, and full-stack AI product engineering.
- I build practical AI systems that connect LLMs, autonomous agents, APIs, and data to solve real problems — from multi-agent research workflows to LLM triage models.
- Director, Product, Strategy & Technology at AskLumenAI, student founder at ViksitHub, and Winner of the National Level AI Summit 2025.
Projects(16)
A LoRA adapter that turns Qwen2.5-1.5B-Instruct into a structured customer-support triage assistant, returning a single JSON object with
emotion,urgency,summary,next_action, andsuggested_reply.- Reported on a 600-ticket synthetic holdout: emotion 55.7% → 99.0%, urgency 19.8% → 94.8%, emotion+urgency exact match 94.5%, valid structured output 99.8%.
- Honest scope: the reply field can invent amounts (~59%) and next-action accuracy is 56.3%, so it is a triage aid with a human in the loop, not automated replies.
- Adapter (~10.6M params) and the reproducible triage dataset are public on Hugging Face.
- LoRA
- PEFT
- Qwen2.5-1.5B
- LLM fine-tuning
- Structured output
- NLP
An autonomous B2B purchasing agent that protects merchant revenue from high-velocity stockouts — built for the Razorpay Buildathon, Track 1 (AI Growth & Agentic Commerce).
- The LLM proposes; a deterministic, AP2-inspired gate disposes. Every money decision is checked against merchant-signed financial boundaries (Ed25519-signed intent → cart → payment mandate chain).
- Runs on Razorpay rails with signature-verified webhooks and idempotent, exactly-once execution. Test/simulated payments only — never real funds.
- Live dashboard and API docs.
- Python
- FastAPI
- LangGraph
- Razorpay
- Ed25519
- React
- TypeScript
An AI software-building operating system that turns a product idea into a PRD, architecture, task breakdown, generated full-stack starter code, a GitHub PR plan, CI/CD workflow, build report, and security review.
- Production-style MVP (independent project): Next.js dashboard with an agent pipeline, pgvector-backed RAG memory, human-approval gates, and audit logs.
- FastAPI backend with JWT auth, SQLAlchemy/Postgres schema, prompt-injection guard, and a Docker Compose + GitHub Actions pipeline.
- Next.js
- TypeScript
- FastAPI
- PostgreSQL
- pgvector
- Redis
- Docker
- GitHub Actions
An AI-powered learning and research assistant. As Director, Product, Strategy & Technology, I lead product and the technical direction.
- Multi-agent research workflows that plan, retrieve, and synthesize across sources.
- A research/personalization layer with rolling memory so answers adapt to each user over time.
- Academic-source integrations (arXiv, Crossref, Semantic Scholar, OpenAlex) with APA/IEEE/Harvard/BibTeX/RIS citations.
- Multi-agent
- RAG
- LLMs
- Personalization
- FastAPI
- Full-stack
Experience
AskLumenAI
- Location
- Pune, India
- Location type
- (Remote)
- Employment status
- Current
- Lead product, strategy, and technology for an AI-powered learning and research assistant.
- Designed multi-agent research workflows that plan, retrieve, and synthesize findings across sources.
- Built a research/personalization layer and rolling memory so responses adapt to each user over time.
- Integrated academic sources (arXiv, Crossref, Semantic Scholar, OpenAlex) with citation formats like APA, IEEE, Harvard, BibTeX, and RIS.
- Own the full stack end to end, from LLM orchestration and retrieval to the shipped product surface.
- Python
- LLMs
- Agentic AI
- RAG
- LangGraph
- FastAPI
- PostgreSQL
- Full-stack
ViksitHub
- Location
- Pune, India
- Location type
- (Remote)
- TypeScript
- Next.js
- React
- Node.js
- Tailwind CSS
- PostgreSQL
- Product
- Full-stack
- Entrepreneurship
- Product Strategy
- Business Development
Xerxez Solutions
- Location
- Remote
- Location type
- (Remote)
- Python
- LLMs
- Generative AI
- NLP
- RAG
Education
- Python
- C++
- Java
- SQL
- DSA
- Machine Learning
- Deep Learning
- DBMS
- Software Engineering
Stack
Research(1)
Hardware-Level Defenses Against Side-Channel Attacks: Mitigating Power and Timing Vulnerabilities in Cryptographic Chips
Amrapali Chavan, Om Patil, Krushna Renuke, Yash Raut, Sudarshan Narhare
- Venue
- International Journal of Creative Research Thoughts (IJCRT)
- Volume
- Vol. 14, Issue 5, pp. a560–a565
- Published
A review of hardware-level techniques for defending cryptographic chips against power and timing side-channel attacks — WDDL, hardware masking, constant-time design, and clock randomization — compared on effectiveness, area/performance overhead, scalability, and residual leakage.
Recognition(7)
- Type
- Award
- Prize
- Winner
- Awarded in
- Received in Grade
- National
- Type
- Award
- Prize
- Runner-up
- Awarded in
- Received in Grade
- National
Complete Data Science, Machine Learning, Deep Learning & NLP Bootcamp
- Type
- Certificate
- Issued by
- Udemy
- Issued on
Introduction to Generative AI
- Type
- Certificate
- Issued by
- Google Cloud
- Issued on
Introduction to Large Language Models
- Type
- Certificate
- Issued by
- Google Cloud
- Issued on
Introduction to Responsible AI
- Type
- Certificate
- Issued by
- Google Cloud
- Issued on
Contact
Support
All of my open-source work — components, blocks, and projects — is free to use. If it saved you time, you can support the work or star the repositories.
Components(44)
Blocks(14)
Open Source 01
An open source section with authored pull requests and repositories.
Support 01
A support call-to-action with sponsor and star actions.
Not Found 01
A 404 page with a playable brick breaker game.
Social Proof 01
A social proof section with a logos carousel.
Social Links 01
A social links section with a lined grid layout.
Metrics 01
A metrics section with a line chart.