I build autonomous AI systems from scratch in Python — no LangChain, no AutoGen, full control over every layer.
My work spans RAG pipelines with ChromaDB and SentenceTransformer, cost-aware LLM orchestration using Gemini API and Groq, deterministic query routing, and modular agent architecture. I also have hands-on production engineering experience — shipped 3 enterprise web applications during my internship, improving workflow efficiency by 30%.
"I don't use frameworks as a crutch — I build from scratch so I actually understand what's happening."
| Project | Description | Status |
|---|---|---|
| 💰 Finance Agent (CLI) | Standalone CLI finance agent powered by Groq — 8 natural-language commands, per-category budget tracking, fully offline. Built on Expense Tracker's OOP foundation. | 🔄 In Progress |
| 🎯 Career AI Agent | Upload resume + job link → get job fit analysis, resume improvements, interview tips & coding questions. Phase-wise execution pipeline. | 🚧 Planning |
| 🔬 Research Agent | Autonomous AI agent with human-in-the-loop control for research tasks and structured output generation. | 🚧 Planning |
| 📄 hArI | RAG-based PDF insights system — upload PDFs via Streamlit UI, ask questions, get grounded answers. Pure Python + ChromaDB + Groq. | 🚧 Planning |
Production-grade autonomous agent built entirely from scratch in Python.
- 4-layer pipeline: cache → pattern router → RAG → LLM
- ~80% of queries resolved with 0 LLM calls via deterministic routing
- ChromaDB + SentenceTransformer RAG with ~30ms semantic search latency
- Planner → Validator → Executor → Responder structured pipeline
- Per-session cost ~$0.0005 with daily quota enforcement
- Stack: Python 3.11+, Gemini API, ChromaDB, SentenceTransformer, DuckDuckGo Search, Open-Meteo
Standalone CLI finance agent powered by Groq — switched from Gemini after hitting quota limits, found better speed and reliability for this use case.
- 8 natural-language commands: add, update, delete, summaries, budgets, dashboards
- Zero code duplication — reuses Expense Tracker classes (~60% dev time reduction)
- Config-driven per-category monthly budget system
- Fully offline storage — no external dependency beyond Groq API
- Stack: Python, Groq API, OOP, JSON storage, CLI
Multi-user expense tracker with secure auth and offline JSON storage.
- Handles 1,000+ records efficiently without a database
- 40% code complexity reduction via OOP design patterns
- Daily/monthly analytics with real-time dashboards
- Foundation reused entirely for Finance Agent — zero duplication
Modular Python + Rich CLI system for full café operations.
- 9 independent modules: auth, menu, orders, analytics, session management
- Secure password hashing/salting + 30-min auto session timeout with 2-level warnings
- 7+ operational features across admin & customer portals
- Full order lifecycle + daily/monthly analytics dashboards
 Stateful multi-agent workflows |
 Human-in-the-loop systems |
 AI app interfaces |
