Senior Director of Engineering with 13+ years spanning full-stack, distributed systems, and enterprise AI platform engineering — a hands-on builder with deep expertise in LLM integration, agentic orchestration, and real-time streaming architectures.
- Designed a three-stage agentic pipeline: Context Preparation → LLM Inference → Tool Invocation Loop with iterative reasoning across 6+ LLM providers (OpenAI, Anthropic, Gemini, Ollama, xAI Grok, Cohere)
- Real-time SSE/WebSocket streaming with step-by-step execution visibility and live token-by-token LLM output
- AES-256 encrypted vault for API key injection at runtime; prompt injection detection covering indirect attacks, hyperbole injections, and payload-splitting
- DAG execution engine with 40+ node types: AI Agent, Webhook, Cron/Schedule, Code Execution, Router, Loop, Parallel Executor, Join, Document Extractor
- Parallel branches, conditional logic, error routing, retry policies, versioning, rollback; event-driven internal message bus
- Pluggable vector DB layer: PostgreSQL+pgvector, Qdrant, Redis; ingestion for PDF, DOCX, XLSX, web pages, Google Drive, OneDrive with GPU-accelerated OCR (Docling/FastAPI)
- Standalone MCP server (Go) exposing 100+ tools across 19+ categories: File System, Google Workspace, MongoDB, GitHub, Jira, Splunk, Odoo ERP; RBAC + SSO (OAuth2, Microsoft Entra)
- Autonomous agent reading entire codebases, multi-file edits, terminal execution and browser automation via natural language
- QA Mode: records browser interactions → auto-generates Playwright test suites, cutting manual authoring from days to hours
- Local model fallback (Ollama); smart context condensation; published as VSIX on VS Code Marketplace
- Head of Full-Stack & AI CoE — led hiring, training, capability development for 150+ engineers across 4 enterprise accounts
- P&L ownership, pre-sales, solution engineering; drove platform strategy and architecture governance
- Projects: XIO (AngularJS 2, RxJS/Redux), OCS (React, Socket.IO, TCP distributed scheduler), PD (React 16 migration), Boxafe (React, Python, PyMongo), Aera Skill Workbench (Product Owner)
- Part of core AI/ML team — Bifrost: AI-based UI generation (Sketch/PSD → React apps)
- Built DAC backend (Node.js, Redis, Socket.IO, PM2, Docker) dispatching millions of search requests
- GEARS: High-perf GRID library (Excel-like) for large datasets with D3/NVD3 data viz
- Product Launch Dashboard: CanJS component architecture, NVD3 visualization customization
Explored and implemented core ML algorithms — regression (Linear, Multi-variable; R², RMSE, MAE), classification (Logistic Regression, Decision Trees, Random Forest — 96.67% on digit recognition), and a full ensemble suite: Bagging (90.5%), AdaBoost (85.5%), Gradient Boosting (89%), XGBoost, Voting, Stacking — applied across Iris, Breast Cancer (AUC 0.9976), Titanic, Churn, Heart Disease, and Diabetes datasets. Built a production chatbot (10 intents, Streamlit + Gradio), salary predictor POC, and explored AI search: A*, Alpha-Beta Pruning, Simulated Annealing. Grounded in Inferential Statistics: probability distributions, hypothesis testing, CLT, and Pearson correlation.
Explored Deep Learning — Perceptron and MLP from scratch (XOR via backprop), optimization algorithms (SGD, Mini-batch, Adagrad, RMSProp), TensorFlow & Keras hands-on. In NLP: built preprocessing pipelines (stemming, lemmatization), TF-IDF and Word2Vec/FastText/GloVe embeddings, LDA topic modeling, SMS spam classifier (5,574 messages), NER and POS tagging with NLTK, spaCy, and Gensim.
In Computer Vision: histogram equalization, geometric transforms, frequency-domain filtering; feature detection via SIFT (128-D), Harris Corner, Hough Transform; shape analysis with Chain Code, Active Contours, Connected Components; segmentation (K-means, Fourier Descriptors) — using OpenCV, SciPy, NumPy.