⚡ AI Agent Engineer

Building agents
that actually work.

LLM-driven agents that select and execute tools autonomously via function calling — deployed and tested, not just prototyped. BCA undergraduate at CSJMU, Kanpur (Class of 2027).

View Projects Get in Touch
Capabilities
Technical Skills

Agentic AI / LLM

LangChainCrewAI Claude APIGroq API MCPFunction/Tool Calling Prompt Eng.

Backend / APIs

PythonFlask REST APIsOAuth 2.0 httpx

Testing / Tools

pytestGit/GitHub Render

Also Familiar With

LangGraphFastAPI RAGChromaDB SentenceTransformersDocker
Work
Projects

AI Email Triage Agent

GitHub → Live Demo →
Python · Flask · LangChain · Groq API · Gmail OAuth 2.0 · Apr–Jul 2026
  • Gmail-integrated agent using Groq (Llama 3.3 70B) via LangChain — the LLM autonomously selects from 3 tools (draft_reply, flag_urgent, archive_silently) through function calling instead of hardcoded branching logic.
  • Achieved 93.5% topic / 96.8% urgency / 91.4% both-correct accuracy on a 93-email evaluation set, confirmed on a separate 30-email holdout (90% / 100% / 90%).
  • Migrated the LLM backend from Claude API to Groq after hitting free-tier quota limits, and corrected a UTF-16 BOM encoding bug introduced during the migration.

CrewAI Blog Writing Agent

GitHub →
Python · CrewAI · Groq API · GitHub API · May–Jul 2026
  • 3-agent pipeline (Technical Analyst → Content Writer → Fact Checker) that fetches a GitHub repository's README and metadata live via the GitHub API and generates a fact-checked blog post introducing the project.
  • Stress-tested the Fact Checker against fabricated claims: caught 3/3 blunt fabrications and a subtly distorted statistic (93%→99%), with 100% detection consistency across 3 repeated runs.
  • Diagnosed and fixed a CrewAI/Groq cache_breakpoint incompatibility (Issue #5886) that caused a 400 error on 100% of runs, plus a silent output bug where crew.kickoff() returned only the final task's result.

GitHub MCP Server

GitHub →
Python · MCP SDK · httpx · pytest · GitHub REST API · Jun–Jul 2026
  • MCP server exposing 3 tools (list_repos, list_issues, create_issue) over stdio transport — lets Claude Desktop manage GitHub repositories via natural language.
  • Added network-level error handling (timeouts, connection failures) and GitHub API rate-limit detection; wrote 10 automated tests (pytest + pytest-httpx) covering success paths, error handling, and edge cases — all 10 passing with 0 warnings.
  • Resolved an SSL verification failure by migrating from a deprecated certifi string-based verification call to an explicit ssl.create_default_context configuration.
Background
About

BCA undergraduate at CSJMU, Kanpur (Class of 2027), targeting Agentic AI Engineer roles — with hands-on experience building agentic systems end-to-end, from LLM orchestration to deployment.

Comfortable debugging production issues: API quota limits, provider migrations, SSL failures, and tool-calling errors — not just writing first-pass code.

Currently focused on MCP integrations and multi-agent systems using CrewAI and Groq.

AI Fluency: Framework & FoundationsAnthropic · 2026
Introduction to AIGoogle · Coursera · May 2026
Maximize Productivity With AI ToolsGoogle · Coursera · May 2026
Machine Learning FundamentalsLearna · Apr 2026
Python for AILearna · Mar 2026
Cyber Job SimulationDeloitte via Forage · Apr 2026

Let's build something.

Open to AI Engineer roles, internships, and freelance projects.

anuragtiwari73219@gmail.com GitHub LinkedIn