Top picks from my engineering portfolio.

Mar 2025

MetaLens | Metadata Store Viewer

Distributed metadata exploration solution for schema exploration and analytics across 4+ lakehouse formats directly from S3-compatible object stores without external catalog dependencies.

• Designed a distributed metadata exploration solution that allows schema exploration and analytics on 4+ lakehouse formats like Hudi, Iceberg, Delta, Parquet, directly from S3-compatible object stores without any dependency on external catalog systems.
• Orchestrated a microservices-based solution that uses Node.js and FastAPI to provide parallel data services for efficient processing of 100+ GB of data.
• Optimized a solution that uses DuckDB and Apache Arrow to provide efficient metadata querying, resulting in a 60% reduction in scanning data.

Mar 2026

GraphMind | Financial Memory Assistant

Multi-tenant AI memory assistant with strict user isolation, hybrid retrieval, and production-grade ingestion/reasoning pipelines for financial knowledge workflows.

• Architected a multi-tenant AI memory system with strict user isolation using JWT-based authentication and dual persistence (PostgreSQL + Neo4j), enabling ingestion of conversational data into structured knowledge graphs with automated entity extraction and relationship mapping, ensuring 99.9% data isolation compliance.
• Engineered a hybrid retrieval pipeline combining graph traversal (Neo4j Cypher) and vector search (Milvus ANN), implementing adaptive query strategies and RRF-based ranking to achieve ~45ms average retrieval latency with context-aware multi-signal scoring.
• Developed a production-grade document ingestion and reasoning pipeline supporting multi-format inputs (PDF, images, text) with OCR (PaddleOCR) and LLM-based extraction, enabling fact deduplication, contradiction detection, and citation-aware response generation with audit trails and quality scoring.

Dec 2025

InsightGuard | APT Detection Framework

Real-time APT detection framework with attack lifecycle observability, simulation-driven validation, and adaptive anomaly detection.

• Engineered a real-time APT detection framework leveraging ELK Stack and host/network telemetry (Sysmon, Winlogbeat, Packetbeat), enabling end-to-end visibility across the attack lifecycle (reconnaissance to exfiltration) with 19+ production-grade detection rules.
• Designed a multi-stage attack simulation engine replicating APT36 tactics mapped to MITRE ATT&CK, enabling controlled validation of detection pipelines and improving threat coverage across persistence, privilege escalation, and data exfiltration scenarios.
• Developed an ML-driven anomaly detection system using a PPO-based reinforcement learning agent with SHAP explainability, enabling adaptive detection of C2 traffic patterns and reducing false positives in network anomaly detection workflows.

Nov 2024

OCDetect | OCD Assessment Platform

Full-stack ML-powered diagnostic platform combining behavioral assessments and computer vision signals to generate real-time OCD severity insights.

• Engineered a full-stack ML-powered diagnostic platform with 20+ interactive assessment modules, processing 50+ behavioral and questionnaire metrics to generate real-time OCD severity scores (0-100%).
• Developed a real-time computer vision pipeline using MediaPipe and OpenCV via a Flask microservice, capturing 100+ temporal behavioral signals per session for multimodal prediction quality.
• Designed a scalable backend and authentication system with Firebase and MongoDB for persistent user profiles, secure protected routes, session tracking, and insight dashboards.

Mar 2026

Coldchain Optimization: Logistics Matching Engine

Real-time cold-chain logistics matching platform with geospatial tracking, workflow orchestration, and pricing intelligence across web and mobile experiences.

• Engineered a real-time cold-chain logistics matching engine using a multi-factor scoring algorithm (route overlap, temperature compatibility, capacity, time window, distance deviation) to rank transporter options with confidence scores, enabling optimal shipper-transporter pairing without prior route visibility.
• Designed a distributed request aggregation and workflow system modeling shipper and transporter interactions, supporting multi-parameter matching with eventual consistency across actors and enabling dynamic decision tracking for acceptance/rejection flows.
• Developed an end-to-end tracking and pricing system with geospatial analytics (PostGIS), real-time ETA updates, and cost computation models, delivering interactive shipment visibility across web and mobile platforms using containerized microservices and JWT-secured multi-role access.