🤖 Agentic AI · ⚙️ Backend Development · ☁️ Cloud Engineering (Azure) 📍 Jaipur, India | 🎓 B.Tech (Hons.) CSE — AI/ML, Manipal University Jaipur 💼 Software Engineering Intern @ Grant Thornton Bharat
I'm a final-year CS (AI/ML) student building production-oriented AI systems — tool-calling agents, document-processing pipelines, and backend services on Microsoft Azure.
These days my work lives at the intersection of:
- Agentic AI — agents that actually do things via API tools, not just chat
- Backend engineering — clean, secure REST APIs with FastAPI
- Cloud — shipping and running it all on Azure
I like turning messy, manual workflows into reliable automated ones — and I care about the boring parts (auth, fallbacks, observability) that make AI tools survive contact with real users.
- Languages: Python, Java, SQL
- Agentic AI: Azure AI Foundry, Tool Calling, MCP Servers, Prompt Engineering, OCR
- Backend: FastAPI, REST APIs, JWT / OAuth2, SQLAlchemy, Alembic
- Cloud & DevOps: Azure App Service, Azure Functions, Azure Storage, Entra ID, Azure DevOps, GitHub Actions, App Insights, CI/CD
- Databases & Tools: PostgreSQL, SQLite, Git, VS Code, Sourcetree
- Also familiar with: PyTorch, TensorFlow, Scikit-learn, Pandas, NumPy, LangChain
FastAPI · PyMuPDF · SQLite · ChromaDB · Python
- Multi-agent workflow that turns a company name into a branded Microsoft Word research report by discovering, validating, and ingesting annual reports.
- Fully local, zero-key architecture using SQLite for queuing, ChromaDB for vectors, local sentence-transformers, and a mock LLM by default (with open LLM support).
- Built an event-driven worker system featuring process-wide rate limiting, concurrent section agents, and fallback CSV data injection.
- Engineered a robust SSRF guard with per-hop redirect validation to securely fetch external PDFs and data sources.
Ollama · Qdrant · FastAPI · React (Vite) · Python
- Hand-rolled 4-agent pipeline (Planner → Retriever → Analyst → Critic) that answers questions over code and PDFs with hybrid retrieval and per-claim citations.
- Built a terminal-style React trace UI featuring an execution graph, latency waterfall, and token-by-token streaming over WebSockets.
- Fully local execution (Ollama, Qdrant, SQLite) with incremental, deterministic AST-aware reindexing via xxhash64 diffing.
- Developed runtime model management to seamlessly swap or pull Ollama chat and embedding models through the UI without restarting.
FastAPI · Playwright · BeautifulSoup · SQLite Queue · Python
- Tool for analysts to upload ZIPs of background-check PDFs, triggering an async worker that scrapes linked sources and uses an LLM to classify adverse risk.
- Developed a multi-stage scraping fallback chain (Playwright → BeautifulSoup → ScraperAPI) protected by strict SSRF guards.
- Features stateful diffing for reprocessed jobs, generating Excel workbooks that highlight risk additions and removals.
- Implemented robust queue mechanics including job visibility timeouts, automatic retries, and dead-letter routing using SQLite.
FastAPI · SQLAlchemy · Alembic · Python
- PrivacyOps MVP to discover PII across databases, logs, tickets, exports, and AI prompt workflows.
- Metadata-first discovery that flags risky fields, scores confidence, and recommends remediation — no raw data upload needed.
- Modules for scan findings, DSR tracking, consent logging, and DPDP compliance evidence.
- Developed a local Next.js dashboard to visualize findings, trace compliance gaps, and aggregate technical readiness evidence.
📈 Earlier ML / fintech projects
- Algorithmic Trading System — FinBERT sentiment analyzer, CUDA-accelerated (<100ms latency), low correlation to SPY
- Time Series Forecasting — LSTM for GOOG price prediction with RSI / Bollinger Band features
- Loan Approval Predictor — Scikit-learn classifier (~86% accuracy) surfacing credit history as a key signal
- Microsoft Certified: Azure AI Engineer / Developer Associate (in progress)
- Microsoft Certified: Azure Fundamentals (AZ-900)
- Microsoft: Artificial Intelligence on Microsoft Azure
- Oracle Academy: Database Programming with SQL
- Multi-agent and tool-calling patterns on Azure AI Foundry
- MCP servers for connecting agents to real backend systems
- Making AI tools production-safe: evals, guardrails, and observability
- Better prompts = more reliable agents
- How I cut manual analyst effort with an agentic background-check tool
- Designing resilient extraction pipelines that don't break on weird sources
- Wiring backend APIs into LLM agents as callable tools
- Running AI workloads on Azure without things falling over
- 📬 rishidixit.7404@icloud.com
- GitHub: you're already here :)
"Production is just a prototype that survived enough" — Rishi Dixit (probably while debugging an agent at 2AM)