MOHAMMAD
OWAIS

I'm a recent B.Tech CS graduate from IIIT Jabalpur (CPI 8.1) who works mostly on backend and search systems, and lately builds with LLMs.

At work I design search infrastructure — dedup pipelines, cross-source indexing, cloud migrations. I care about systems that are correct, fast, and honest about their trade-offs. Competitive programming on the side keeps my problem-solving sharp.

• Built AI job summaries with Amazon Bedrock (Claude Haiku 4.5) so advisors can scan roles without opening each listing.
• Architected a cross-source dedup system (facility + role + employment composite key) over 256,000+ records from 7 sources, removing 12,600 duplicates while keeping every job findable per source.
• Leading the Azure→AWS migration — onto OpenSearch, ECS Fargate and S3, including a full rewrite of the query builder from Azure OData to OpenSearch DSL.

• Built the full automated pipeline on Azure for a healthcare job-search platform — scrapers for 4 Japanese portals, preprocessing and indexing into Azure AI Search — scaling it to 145,000+ listings across all 47 prefectures with 6-worker concurrency and a resume layer for network failures.
• Generated AI summaries for the full catalog with Azure OpenAI, running 20 parallel calls and handling content-filter edge cases to reach near 100% coverage.
• Fixed distance search returning no nursing jobs — lifting coordinate coverage from 51% to 100% and results within 10 km of Tokyo from 0 to over 5,400.

A RAG pipeline over PDFs built from scratch — no LangChain, no vector database. Token-aware chunking, sentence-transformer embeddings and cosine search, with page-level citations and a relevance gate that refuses off-topic questions before spending an API call. Includes an eval harness (hit@k, MRR) that runs without an API key.

Semantic job search + RAG assistant over live listings. Gemini embeddings and a FAISS vector store for retrieval, a LangGraph workflow orchestrating retrieval → prompt → generation, and cited answers served over a FastAPI API. Deployed with Docker.

Two-sided platform connecting restaurants' surplus food with nearby NGOs. Real-time orders & chat over Socket.IO, an SVD recommender built from scratch in NumPy plus content-based filtering, and hardened JWT-in-httpOnly-cookie auth closing a Broken-Access-Control gap.

Real-time ambulance management with a GenAI-powered triage chatbot and sentiment analysis for case classification. Dynamic allocation for faster response, and WebSocket notifications pushing live ETA & ambulance details to patients and hospitals.

AI-powered mock-interview platform that generates role-specific questions, runs the session, and gives structured feedback on your answers.

Conversational weather assistant — ask about conditions in plain language and get chat-style forecasts backed by a live weather API.

Fine-tuned DistilBERT to detect SMS smishing at 99.4% accuracy, fixing a 6:1 class imbalance with back-translation. For URLs, benchmarked Logistic Regression vs Random Forest (82.5%) on 549K samples — choosing the Random Forest over a 99% DistilBERT that failed to generalise. Ships with a Chrome MV3 extension that scans page links against the deployed model.

Computer-vision pipeline that detects drowning events from pool footage to trigger early rescue alerts.