Featured Projects
Deep dive into my recent work in AI Engineering and Full-Stack Data Science.
System Architecture
Click diagram to enlarge
Cold Email Personalization AI
AI-Powered SDR Agent for Automated, Hyper-Personalized Cold Outreach.
Key Features
- End-to-End RAG: Generates context-aware emails using Gemini 2.5 Flash, Google Embeddings 001, and Pinecone vector search.
- Resilient Scraping: Uses FastAPI and Playwright to extract clean domain text automatically.
- Mobile-First UI & Web Intents: Features a WhatsApp-style chat interface and bypasses cloud SMTP blocks using secure Gmail Web Intents.
Impact & Results
- Automated Prospecting: Eliminates manual research by instantly scraping and analyzing lead websites in real-time.
- Hyper-Personalization: Crafts highly relevant hooks grounded in actual company data and stored value propositions.
- Zero-Cost Delivery: Replaces fragile SMTP setups with local browser handoffs, guaranteeing 100% deliverability without paid email APIs.
Click diagram to enlarge
Multimodal RAG Assistant
Chat with your PDFs, Images, and Videos in Real-Time.
Key Features
- Multimodal Ingestion: Seamless semantic processing of PDFs, images (Tesseract OCR + Salesforce BLIP), and video (OpenAI Whisper + OpenCV).
- Visual & Audio Intelligence: Generated comprehensive video timelines by combining audio transcription with automated frame-by-frame visual captioning.
- Memory-Aware Chat: Implemented session-level conversational memory via LangChain to maintain deep context during complex, multi-turn analytical queries.
Impact & Results
- Proven Accuracy: Validated the retrieval pipeline via the Ragas framework, achieving a highly rigorous 0.98 Context Precision and 0.96 Answer Relevancy.
- Optimized Cloud Deployment: Successfully deployed on Hugging Face Spaces, resolving Streamlit OOM constraints by managing Linux dependencies within a 16GB RAM container.
- Unlocked "Dark Data": Transformed opaque multimedia assets into instantly searchable, grounded knowledge, completely eliminating LLM hallucinations.
Click diagram to enlarge
FinPulse AI
Real-Time Market Sentiment & Predictive Analytics.
Key Features
- Hybrid Engine: Correlates live stock prices with real-time news sentiment.
- Institutional NLP: Uses FinBERT to score news as Bullish/Bearish accurately.
- 7-Day Forecasting: Deploys Prophet for confidence-bound price predictions.
Impact & Results
- Quantified Sentiment: Converted subjective news into actionable numerical scores.
- Risk Management: Provided reliable 7-day lookahead windows with confidence intervals.
- Unified Dashboard: Consolidated disjointed data streams into one UI.
Click diagram to enlarge
Smart Finance Tracker
AI-Powered Personal Finance Dashboard & Forecaster.
Key Features
- Multi-Model Extraction: Uses Llama 3.3 to parse complex PDF statements.
- Hybrid Categorization: Llama 3.3 auto-labels transactions with intelligent caching.
- Predictive Analytics: Forecasts month-end spending using Linear Regression models.
Impact & Results
- Data Liberation: Automated the pipeline from static PDF to interactive dashboard.
- Cost Efficiency: Reduced API costs via a "Tiered Intelligence" caching strategy.
- Actionable Insights: Enabled "Chat with Data" via RAG for instant financial queries.