At Valentius Kryptix, I built a document question-answering chatbot over a 543-document technical corpus. It uses ChromaDB for retrieval and Google Gemini to answer from retrieved context, with source details returned alongside each answer.
In my Support-Ticket Triage project, I fine-tuned DistilBERT on BANKING77 and reached 92.72% mean test accuracy across three seeds. I also calibrated its confidence and wrapped the model in a FastAPI endpoint with rule-based routing.
For my Delivery ETA Engine project, I used XGBoost to model delivery times, reaching a 3.21-minute test MAE against a 7.58-minute median baseline. I served ETAs and SHAP factors through FastAPI and logged outcomes to PostgreSQL.
I recently graduated in Electronics and Telecommunication Engineering with Honors in IoT. I’m also an IJSREM-published author, with a publication on “GestureBot AI: Smart Surveillance Car.”
