Professional TwinBot AI

Resume-grounded agentic assistant: tool-style prompting + retrieval patterns for accurate, personalized responses.

Live Demo · · Back to Projects

  • Problem: Make a resume “queryable” so recruiters/teams can quickly explore relevant experience via conversation.
  • Approach: Resume-grounded prompting + retrieval-style context injection + lightweight tool-use patterns.
  • Outcome: Deployed an interactive Gradio app on Hugging Face Spaces; supports multiple LLM backends (local + hosted).

Demo implementation of an agentic-style career assistant that uses LLMs and basic tool-enabled workflows to present my professional background conversationally. Built as part of hands-on Agentic AI labs.

The key components of the implementation include:

  • Context ingestion: loads resume + curated summary as the source-of-truth for responses.
  • Grounding strategy: uses a condensed resume context window to stay within token limits.
  • LLM backends: supports local and hosted models (Ollama / OpenAI-compatible API) behind a simple interface.
  • Agentic patterns: structured prompting, tool-style routing, and multi-step response assembly for better relevance.
  • UI & deployment: Gradio chat interface deployed on Hugging Face Spaces for easy sharing and testing.

Engineering Notes

  • Truth source: the resume text is treated as the ground truth; responses are expected to stay within that context.
  • Reliability: prompt structure encourages quoting/pointing to the underlying resume sections rather than inventing details.
  • Safety & privacy: avoids collecting sensitive user data; limits outputs to professional info included in the provided documents.
  • Deployment tradeoffs: Hugging Face Spaces + Gradio keeps iteration fast, but model latency depends on the selected backend.
  • Next improvements: add a small eval set, track grounding rate, and add refusal behavior for out-of-scope questions.

Limitations

  • Not a general career advisor; designed to answer questions specifically about my background and projects.
  • Quality depends on resume clarity and model choice; complex questions may require follow-up prompts.