Ojaswi Barotia
AI & Full Stack Developer | Building Smarter Digital Product
New to Contra
Ojaswi is building their profile!
AI Voice Agent: Autonomous Calling & Support
A US-based service provider needed an ultra-low-latency autonomous AI Voice Calling Agent to handle inbound customer qualification, resolve routine queries, and automatically book calendar appointments 24/7 without manual front-desk staff.
Key Challenges:
High conversational latency causing unnatural pauses and awkward caller overlap.
Managing interruptions dynamically when a user speaks while the AI is responding.
Live calendar slot booking and syncing caller records instantly with the client's CRM.
Technical Implementation & Solution:
• Telephony & Audio Pipeline: Configured bi-directional WebSocket audio streams linking Twilio Voice directly to FastAPI backends.
• Ultra-Low Latency Voice Stack: Integrated Deepgram Nova-2 for real-time speech-to-text (STT), OpenAI GPT-4o-mini for sub-second intent reasoning, and Cartesia / ElevenLabs for natural, emotional voice synthesis.
• Interruption Handling: Built a voice activity detector (VAD) to instantly pause AI speech playback when the caller speaks mid-sentence.
• Tool Execution & Integrations: Integrated dynamic function calls directly into Google Calendar and HubSpot CRM to lock meeting slots during live conversations.
Results & Impact:
• Reached an average response latency of under 650ms, producing natural human-like voice flow.
• Automated 70% of inbound calls, cutting front-desk operating expenses significantly.
• Successfully scheduled qualified leads directly into CRM with zero human intervention.
1
1
71
AI Fraud Detection: Real-Time Risk Engine
A digital payments startup processing over 100,000 daily transactions needed a high-throughput, low-latency AI fraud detection engine to intercept fraudulent charges, account takeovers (ATO), and identity theft in real time without creating friction for legitimate customers.
Key Challenges:
Reducing high false-positive rates that caused cart abandonment and user churn.
Evaluating complex behavioral features and payment metadata in under 60 milliseconds.
Adapting to rapidly shifting fraud tactics without requiring full model retraining.
Technical Implementation & Solution:
• Real-Time Feature Pipeline: Engineered an ultra-fast streaming pipeline using Apache Kafka, Redis, and FastAPI to compute transaction velocity and device fingerprints on the fly.
• Hybrid Machine Learning Architecture: Deployed an ensemble model combining XGBoost, Isolation Forests, and Graph Neural Networks (GNN) to uncover hidden mule-account rings.
• Explainable AI (XAI): Integrated SHAP values into the decision engine, providing compliance analysts with immediate human-readable reasons for flagged transactions.
• Automated Webhook Routing: Created dynamic webhooks to instantly trigger Step-Up 2FA, soft declines, or outright blocks based on real-time risk scores.
Results & Impact:
• Slashed fraudulent chargeback losses by 43% in the first quarter.
• Reduced false-positive transaction declines by 28%.
• Achieved sub-45ms average inference latency across peak payment volumes.
1
2
100
AI E-Commerce Personalization Engine
A multi-brand online retail store wanted to replace generic homepage banners and static "related items" grids with an intelligent, real-time AI recommendation engine to boost cart value, conversions, and customer retention.
Key Challenges:
Cold-start problem for newly launched products with zero sales history.
Generating sub-50ms dynamic product feeds during peak flash-sale traffic spikes.
Tracking complex visitor browse events without slowing down frontend page loads.
Technical Implementation & Solution:
• Real-time Tracking: Built an asynchronous event-streaming pipeline using Kafka and FastAPI to record clickstreams, cart additions, and dwell time.
• Hybrid Recommendation Logic: Combined collaborative filtering with multi-modal vector embeddings (CLIP) to analyze both user purchase patterns and product visual styles.
• Dynamic Visual Search: Implemented image-similarity search via Qdrant, allowing shoppers to upload photos and immediately discover matching catalog apparel.
• Headless API & Shopify Integration: Exposed lightweight REST & GraphQL endpoints that integrate seamlessly with Shopify Liquid and Next.js frontends.
Results & Impact:
• Increased average order value (AOV) by 22% within 45 days.
• Lifted product click-through rates (CTR) on personalized widgets by 34%.
• Handled 10,000+ concurrent requests with sub-40ms response latency.
1
82
Enterprise RAG AI: Private Doc Intelligence
A fast-growing B2B software firm needed a secure, enterprise-grade Retrieval-Augmented Generation (RAG) system to query thousands of confidential internal documents (legal contracts, technical architecture specs, Notion pages, and employee SOPs) in real time without third-party data leakage.
Key Challenges:
High hallucination risk and irrelevant context retrieval using standard keyword search.
Managing multi-format files (scanned PDFs, tables, markdown) with strict access control based on user hierarchy.
Technical Implementation & Solution:
• Engineered an automated ingestion pipeline in Python with LlamaIndex and LangChain for semantic chunking and metadata preservation.
• Integrated a hybrid vector search architecture pairing dense embeddings (OpenAI text-embedding-3) with sparse BM25 indexing in Qdrant.
• Added a cross-encoder re-ranking layer (Cohere Re-ranker) to prune low-relevance chunks, reducing API token costs by 40%.
• Built robust Role-Based Access Control (RBAC) via FastAPI endpoints, ensuring departmental data isolation.
Results & Impact:
• Dropped internal information retrieval time from 20 minutes to under 4 seconds.
• Attained 98% factual precision with traceable in-line page citations.
• Full on-prem Docker deployment ensuring 100% private data security.
1
2
120