HI, I'M
AI Engineer focused on Agentic AI, automation, and rapid MVP development. Experienced in building enterprise-grade workflows using multi-agent orchestration, hybrid retrieval-augmented generation (RAG), and LLM-powered automation. Comfortable working independently across the full stack data pipelines, model training, ontology design, and backend APIs and skilled at using AI coding agents (Claude Code, Gemini, Codex, Local Models) to move from idea to working prototype in days, not weeks.
Engineered for Accuracy & Scalability
Rigorous evaluation across production retrieval architectures, predictive classification, and deep computer vision systems.
Deterministic Agentic AI & Hybrid Retrieval
Interactive execution flows across my flagship architectures on GitHub (@sanil2050). Watch live data packets stream across routing hubs, dual-branch vector retrievers, and self-auditing critic loops.
Deterministic Planning & Rewriting
User queries are authenticated with JWT and RBAC permissions, then rewritten with conversational context to extract deterministic intent.
Dual-Branch Retrieval & Verification
Dispatches parallel retrieval across pgvector semantic chunks and GraphRAG knowledge triples, fused via RRF and cross-encoder reranked before Nemotron synthesis.
Integrated Toolchain & Storage
Wired across production vector stores, relational DBs, LLM reasoning APIs, and live web retrieval endpoints for zero-latency execution.
Clinical Risk Prediction
XGBoost with SHAP explainability on clinical risk benchmarks (98.19% ROC-AUC).
Featured Projects
Real-world systems spanning autonomous Agentic AI pipelines, deep learning vision models, and enterprise data analytics.
Enterprise RAG Platform
Built a production-oriented RAG platform for secure Q&A over internal company knowledge with JWT authentication and RBAC-based permission filtering. Designed query rewriting using conversation history, dispatching an Agent Planner across Document RAG (hybrid dense + sparse retrieval) and GraphRAG branches (bounded to 3 steps). Fused with Reciprocal Rank Fusion (RRF) and cross-encoder reranking, enforcing evidence-grounded answers with Nemotron.
Hybrid RAG Pipeline
End-to-end PDF ingestion pipeline with pypdf, text cleaning, and recursive chunking (~500 chars, 50 overlap). Generated 384-dimensional dense vector embeddings with HuggingFace all-MiniLM-L6-v2 and fitted a BM25 sparse encoder into a Pinecone hybrid index. Features batched upserts and a PineconeHybridSearchRetriever (top-k=5) streaming grounded answers via Google Gemini 2.5 Flash with strict page-level source citations.
Multi-Agent AI Research System
Autonomous 4-agent collaborative pipeline: Research Agent queries Tavily Search API for ranked web summaries, Reader Agent isolates clean content with BeautifulSoup, Writer Agent synthesizes structured reports with citations using Gemini 2.5 Flash, and Critic Agent audits reports with numerical scores (1–10) and actionable feedback. Sequenced with Streamlit web UI (live execution logs) and CLI mode.
Experience & Applied Work
Hands-on technical leadership developing intelligent agent systems, multi-modal vision models, and enterprise data analytics.
AI Engineer & Data Analyst
Applied Research & Independent Engineering- Built an enterprise-grade, production-oriented RAG platform enforcing JWT authentication & RBAC filtering, query rewriting, an Agent Planner routing across Document RAG & GraphRAG (bounded to 3 steps), RRF fusion, cross-encoder reranking, and grounded answers with Nemotron via FastAPI and PostgreSQL + pgvector.
- Engineered an end-to-end Hybrid RAG pipeline combining dense vector embeddings (HuggingFace all-MiniLM-L6-v2) and BM25 sparse retrieval into a Pinecone hybrid index, ensuring strict page-level source traceability with Google Gemini 2.5 Flash.
- Orchestrated autonomous multi-agent systems (LangChain, Tavily API, BeautifulSoup, Gemini 2.5 Flash) featuring Research, Reader, Writer, and self-auditing Critic agents with Streamlit web UI (live logs) and CLI execution modes.
- Conducted deep learning computer vision research on mammography classification benchmarking YOLOv11 & YOLOv26 on CBIS-DDSM & VinDr-Mammo; trained custom YOLOv8 object detection to 92% accuracy with ONNX export.
- Developed end-to-end data analysis workflows in Python/Pandas, Scikit-learn, XGBoost (with SHAP explainability), and created business intelligence reporting using Tableau, Power BI, and Microsoft Excel.
Specialized Engineering Capabilities
Bridging autonomous Agentic AI systems, enterprise vector retrieval, computer vision, and scalable production APIs.
Agentic AI &
Multi-Agent Systems
Designing autonomous multi-agent orchestration pipelines using LangChain, Model Context Protocol (MCP), and structured critic-auditing loops for rapid MVP turnaround.
Enterprise &
Hybrid RAG
Building production-grade retrieval platforms blending dense semantic vectors, BM25 sparse search, GraphRAG, and RBAC security on Pinecone and PostgreSQL pgvector.
Data Analysis &
BI Dashboards
Transforming complex datasets into actionable business intelligence using Pandas, NumPy, Microsoft Excel, Tableau, and Power BI with exploratory data analysis.
Computer Vision &
YOLO Fine-Tuning
Developing high-accuracy detection and medical classification systems using PyTorch, Ultralytics YOLOv8/v11/v26, and ONNX export for low-latency inference.
Scalable Backend &
APIs
Architecting resilient server-side architectures and clean RESTful APIs with FastAPI and Django, complete with Swagger documentation and JWT security.
Explainable AI &
Diagnostics
Surfacing interpretable model predictions using SHAP (Shapley Additive exPlanations), class-balance optimization (SMOTE), and rigorous metric benchmarks.
Frequently Asked Questions
Answers to common questions regarding my development workflow, tech stack, and engagement availability.
What is your turnaround time for a working AI MVP or prototype?
How do you ensure zero hallucination in RAG deployments?
Are you available for full-time engineering roles or contract work?
Can you work with proprietary on-premise data and custom fine-tuning?
Let's Build Something Impactful
Looking for an AI Engineer, Data Analyst, or autonomous pipeline collaborator? Drop a note or reach out directly.
Send a Direct Message
Fill in your details below to launch your default mail client with prefilled context.