🎓 Master's in CS @ University at Buffalo   |   🤖 AI/ML Enthusiast & SWE Explorer
🔗 Email • LinkedIn

I build intelligent systems — from scientific RAG pipelines to medical diagnostics,
and love turning deep learning + Software engineering into real-world applications.

• 🚀 I enjoy building production-grade AI systems and developer-focused tools with measurable real-world impact
• 💡 Passionate about the intersection of LLMs, RAG, multimodal learning, and software engineering
• 🧰 Equally confident in training deep models or scaling fast APIs and databases
• 💼 Open to roles in: AI/ML Engineering, Fullstack Development, Software Engineering, and Applied Research
• 📫 Reach out:
  • LinkedIn
  • 📧 ashrujgautam4@gmail.com

Built an advanced Retrieval-Augmented Generation system for scientific PDFs — integrating section classification, citation graph analytics, figure detection, and local LLM generation (Mistral 7B).

• 💡 Used Grobid, LayoutLM, ChromaDB, and fine-tuned BERT
• 🔗 Full hybrid vector + keyword retrieval system with prompt optimization
• 📚 F1 Score: 99.54% (Reference Parsing), 79.94% (Section Classification)

AI diagnostic system combining X-ray imaging, clinical NLP, and tabular risk analytics into a unified medical assistant with prescription guidance.

• 🩻 DenseNet121 (CheXpert) + BioBERT + MLP Ensemble (Kidney, Heart, Diabetes)
• 🤖 Integrated with LLaMA for contextual recommendations
• 💻 Packaged with PyQt5 GUI and local data fusion

Production-grade PostgreSQL analytics engine with stored procedures, triggers, indexing, and planned LLM augmentation for feedback analysis.

• 🧠 11+ table schema, optimized queries, and stored functions in PL/pgSQL
• 📈 Real-time business metrics: customer behavior, delays, seller ranks
• 🧠 Extending with Zephyr LLM to auto-summarize trends & complaints

Custom YOLO variant optimized for occluded fruit detection in retail scenarios. Compared performance against SSD, Faster R-CNN.

• 🖼️ Trained on MinneApple + COCO with resolution-aware tuning
• 🚀 Real-time object detection via optimized bounding box anchors

Built a semantic image search engine that lets users query with text or images and returns top-matching results using precomputed embeddings.

• 🧠 Uses OpenAI CLIP + FAISS for vector similarity
• 🖼️ Visual & text-to-image search in real-time

Languages:

Frameworks & APIs:

AI & ML:

Data & Infra:

“Code is the closest thing we have to magic — I just make sure mine solves the right problems.”