1. Semantic Search Engine using Embeddings
Tech: Python, Hugging Face, Sentence Transformers, FAISS, FastAPI
- Implemented Sentence-Transformers–based embeddings for semantic similarity search.
- Designed a FAISS vector index enabling low-latency retrieval across 10k+ documents.
- Developed FastAPI endpoints for indexing, embedding generation, and querying.
- Optimized memory usage and vector store performance, improving efficiency by 20%.
2. OCR + LLM Document Intelligence System
Tech: Python, Tesseract OCR, Hugging Face Transformers, LLMs
- Built an end-to-end OCR → LLM pipeline for structured document extraction.
- Designed prompt templates for entity extraction, summarization, and classification.
- Achieved 92% field extraction accuracy across invoices, IDs, and forms.
- Improved OCR pre-processing (denoising, thresholding), increasing clarity/signals significantly.
3. n8n Workflow Automation Suite
Tech: n8n, Node.js, REST APIs, Cloud Functions
- Created 6+ automation workflows for ETL, notifications, LLM triggers, and reporting.
- Developed custom n8n nodes supporting async REST API orchestration.
- Reduced manual effort by 80% through scheduled and event-driven flows.
- Added robust auto-retry logic, error propagation, and monitoring alerts.
4. Scalable REST API Platform
Tech: Node.js, Express.js, MongoDB, JWT
- Developed a modular backend with controller–service–middleware architecture.
- Implemented JWT authentication, rate-limiting, request logging, and global error handling.
- Optimized database queries and in-memory caching, lowering response times to .
- Integrated CI checks for linting, unit testing, and code coverage enforcement.
5. CI/CD Pipeline Template
Tech: Docker, GitHub Actions, AWS EC2
- Designed reusable CI/CD templates automating build → test → deploy workflows.
- Containerized services using Docker for consistent, reproducible environment setups.
- Enabled zero-downtime deployments using rolling update strategies on EC2.
- Reduced deployment time from 20 minutes to under 3 minutes with parallelized steps.