I'm a final year Information Science Engineering student at BMS Instituteof Technology & Management, Bengaluru, working
across three areas: backend systems, full-stack web development, and applied GenAI/LLMtooling.
On the backend, I build REST APIs, auth flows, and rate-limited services withDjango REST Framework and FastAPI, backed by
PostgreSQL and Redis. On the frontend, I pair those APIs with React interfaces. And on the AI side, I buildRAG/RAT
pipelines — retrieval-augmented pipelines using LangChain, ChromaDB,and Groq's Llama 3.1 models — as a way to apply LLMs to
real problems likequiz generation and log analysis.
I'm currently deepening my DSA fundamentals (Trees, Graphs, DP) alongside contributing to open source through SSoC '26
and actively buildingfull-stack + AI projects end-to-end: API design, data modeling, and frontenddelivery.
🎯 Open To: Internships · Full-Time SDE Roles · Open Source Collaboration · Freelance Projects
Languages
Backend & APIs
Frontend
Databases & Caching
AI / LLM Application Development
Cloud, DevOps & Tooling
| Area | Level | Details |
|---|---|---|
| Backend Development (Django/FastAPI) | ●●●○○ | Built REST APIs, JWT auth with token rotation, and Redis-backed rate limiting across multiple projects |
| Full-Stack (React) | ●●●○○ | Ship React/Vite frontends paired with self-built backends; comfortable with Tailwind and component architecture |
| RAG/RAT Pipelines | ●●●○○ | Built a custom multi-pass retrieval engine (LogAnalyzer) and a multi-source RAG pipeline (MindArc) using LangChain + ChromaDB |
| LLM Inference (Groq / Llama 3.1) | ●●●○○ | Integrated Llama 3.1 8B for quiz generation and log reasoning through structured prompt chaining |
| DSA / Problem Solving | ●●●○○ | 200+ LeetCode problems solved, contest rating ~1490; Trees, Graphs, DP in progress |
| Vector Databases & Embeddings | ●●○○○ | ChromaDB with per-session collections and metadata filtering; sentence-transformers for semantic search |
🧠 LogAnalyzer — RAT-Based Incident Reasoning Engine
A RAT (Retrieval-Augmented Thinking) based incident reasoning engine that ingests logs from multiple microservices and runs iterative multi-pass retrieval to produce root-cause analyses — removing the need for manual log correlation during outages.
| Stack | FastAPI · LangChain · ChromaDB · Groq API (Llama 3.1) · React.js |
| Scale | 3 log format parsers · multi-service ingestion |
| Performance | 3-pass iterative retrieval loop over vector store |
| Security | Session-isolated, metadata-aware retrieval collections |
| Impact | Eliminates manual log correlation across microservices during outages |
| Repository | github.com/Tirthpanchori/LogAnalyzer · Live |
Built a custom RAT engine that correlates logs across services through iterative retrieval, producing root-cause analyses with confidence scores. A metadata-aware retrieval layer on ChromaDB uses per-session collections and Unix-timestamp filtering, with purpose-built parsers for three log formats and time-windowed chunking to sharpen retrieval precision.
📚 MindArc — AI-Powered Quiz Generation Platform
An AI quiz-generation platform that turns YouTube videos, PDFs, and raw text into structured quizzes with real-time scoring — built to cut educators' quiz-creation and grading time.
| Stack | Django REST Framework · React.js · PostgreSQL · Groq API (Llama 3.1) |
| Scale | 15+ RESTful endpoints · 3 input source types |
| Performance | Sub-50ms response on indexed quiz retrieval queries |
| Security | JWT authentication with access/refresh token rotation |
| Impact | ~70% reduction in quiz-creation time · ~90% reduction in manual grading |
| Repository | github.com/Tirthpanchori/MindArc · Live |
Integrated Groq's Llama 3.1 8B with structured prompt chaining across three input sources — YouTube, PDF, and raw text — to auto-generate structured quizzes. Backed by 15+ RESTful endpoints, JWT auth with rotating tokens, and indexed PostgreSQL queries, alongside an automated evaluation system that removes most manual grading work.
🔗 URL Shortener with Rate Limiter
A production-style URL shortening backend with Base62 encoding, custom slugs, and configurable link expiry, built to handle abuse at scale rather than just redirect traffic.
| Stack | Django REST Framework · PostgreSQL · Redis |
| Scale | Sliding-window rate limiting across creation & redirect endpoints |
| Performance | O(log N) complexity via Redis ZSET |
| Security | 10 req/60s creation · 60 req/60s redirects, abuse-resilient |
| Impact | Replaced a naive in-memory limiter that couldn't survive server restarts |
| Repository | github.com/Tirthpanchori/URL-Shortener-and-Rate-Limiter |
Engineered a Redis ZSET–based sliding-window rate limiter with O(log N) complexity, along with Base62 encoding, custom slugs, and configurable expiry — replacing a fragile in-memory approach with a persistent, abuse-resistant design.
🌱 Open Source Contributions — Summer of Code '26
| Project | Stack | Contribution |
|---|---|---|
| FreshScanAi | FastAPI · slowapi · pytest | Implemented per-user rate limiting on ML inference endpoints (20 req/min scan, 100 req/hr global), keyed by Supabase user ID with IP fallback, JSON 429 responses, full pytest coverage |
| Agronavis AI Farm Assistant | FastAPI · Supabase | Replaced hardcoded disease dictionaries with dynamic Supabase queries, migrated 86 entries, fixed disease-wiki endpoint to query by class_key with UUID fallback |
| ai-agent-automation | React · Tailwind CSS v4 | Improved mobile responsiveness across dashboard and sidebar with animated hamburger drawer, backdrop overlay, and Tailwind v4 CSS variable shorthand |
| Recognition | Details |
|---|---|
| 🏆 LeetCode — 200+ Problems Solved | Contest Rating: 1490 · Sliding Window, HashMaps. Trees, Graphs, DP in progress |
| 🌱 Open Source Contributor — Summer of Code '26 | Merged PRs across 6 independent open-source repositories |
Generated via Platane/snk — requires the snake GitHub Action enabled on the profile repo
current_focus:
learning:
- Advanced Distributed Systems & System Design
- Graphs, DP, and advanced algorithmic patterns
building:
- Production-grade RAG/RAT pipelines with LangChain + ChromaDB
- Backend services with strict latency & abuse-resistance guarantees
exploring:
- Multi-agent LLM orchestration
- Vector search optimization at scale
open_to:
- Internships
- Full-Time SDE Roles
- AI/ML Engineering Roles
- Open Source Collaboration
- Freelance Projects
"Systems that hold up under real constraints matter more than systems that only work in demos."