Hi, I'm Tulika Das

Problem: Customer support teams struggle with high complaint volumes, mixing urgent issues with trivial queries and causing missed escalations.

Solution: Built a multi-agent system using LangGraph, Llama-3, and Gemini that auto-routes complaints by urgency, loops in human managers for high-severity cases, and sends automated email resolutions.

Problem: Manual research is slow and fragmented — searching multiple queries, reading articles, and synthesizing findings takes hours.

Solution: Built a ReAct agent using LangGraph and Groq that autonomously searches DuckDuckGo across multiple angles and compiles findings into a structured 6-section research report from the CLI.

Problem: Existing safety tools for women are reactive — alerting after danger occurs rather than providing proactive, situational guidance.

Solution: Built an AI safety advisor using Llama-3.3-70b and LangChain with Chain-of-Thought reasoning that delivers age-specific safety tips, self-defense techniques, and relevant video resources.

Problem: Deploying LLMs in production requires more than API calls — teams need rate limiting, cost control, auth, and fallback logic to run reliably at scale.

Solution: Built a scalable LLM backend using FastAPI with authentication, per-user rate limiting, token cost accounting, and provider fallback logic.

Problem: Most LangChain tutorials treat memory as a black box, leaving developers confused about what actually persists and what gets lost between sessions.

Solution: Built a focused system demonstrating both memory types — in-memory short-term context via LangChain and JSON-based long-term persistent storage, with daily reminders via GitHub Actions to log learnings.