Aishwarya Y L Resume Q&A
Upload a text PDF, DOCX, or TXT resume, or paste its text. Ask specific questions about the resume. The app retrieves excerpts and accepts generated answers only when their normalised text appears in the excerpt. This check does not establish that an answer correctly addresses the question.
Deploy the chatbot
Create a Gradio Space on Hugging Face. Upload the contents of this folder directly into the repository root, including app.py, requirements.txt and this README.md. Do not upload the outer folder or the ZIP itself.
This source package does not contain trained weights. Until you supply them, the app loads the original google/flan-t5-small model and displays that status. This is not the personalised fine-tuned model.
Train and use the same model as the notebook
Open Aishwarya_Y_L_Resume_Bot_Colab.ipynb in Colab. Install dependencies, restart the session, and run the training and save cells. Use a GPU when available. The notebook creates the real weights, tokenizer, configuration and measured evaluation outputs. No training results are supplied in this package.
Extract the generated model ZIP and upload its aishwarya_yl_resume_qa_finetuned folder beside app.py in the Space. Alternatively, upload the contents of that folder to a Hugging Face model repository, then set the Space variable MODEL_REPO to that repository's actual ID. For a private model repository, configure HF_TOKEN as a Space secret. Never put tokens in source files.
The model repository should contain the generated config.json, model.safetensors (or generated weight shards), tokenizer files, README.md, metrics.json and test_answers.csv. Use the files actually saved by the notebook; do not create empty weight files. A model repository shares weights; the Space serves the chatbot.
Local use
Install Python 3.10 or later, then run:
pip install -r requirements.txt
python app.py
For script training, install requirements-training.txt and run python train.py from this directory. Script training has the same synthetic data, person-level split, five training epochs and held-out comparison as the notebook. Plots may require an appropriate local plotting backend. ipython is needed for the display function.
Data and limitations
The CSV contains 30 explicitly fictional profiles. No personal resume facts for Aishwarya Y L have been invented or included. Uploading a resume supplies inference context and does not retrain the model. Training uses 20 people, validation uses 5, and testing uses 5. Exact text match is reported only when training and evaluation run. These repeated synthetic records do not establish performance on real resumes.
Retrieval uses English TF-IDF and can miss relevant text. The model can produce incorrect answers; the text-support check can reject valid paraphrases or accept irrelevant copied text. Scanned PDFs require OCR before use. Uploaded documents and chat histories are not deliberately persisted by this code. Review hosting and access settings before uploading sensitive resumes.
Configuration reference: https://huggingface.co/docs/hub/en/spaces-config-reference