A RESTful backend for managing the job hunt: applications, statuses, and follow-ups, structured the way a production API should be.

• Built with C#, ASP.NET Core, and Entity Framework Core on SQL Server, using LINQ for filtering and status reporting.
• Layered architecture with DTO-based request validation, JWT authentication, and global error handling.
• Documented with Swagger/OpenAPI so every endpoint can be tried from the browser.

A mobile calorie tracker built for desi cuisine (karahi, biryani, roti, daal), where generic trackers fall short on dishes and portion sizes.

• Published on Google Play and improved through post-release updates; since retired to focus on newer projects.
• Food database of Pakistani dishes with portion-based logging (per roti, per plate, per bowl) instead of gram-only entries.
• Daily calorie tracking with a simple, fast logging flow built for everyday use.

Built a low-latency WebRTC audio voice transport agent using LiveKit, Whisper (STT), and OpenAI (TTS) for hands-free diagnostics on active Kubernetes cluster environments.

• Orchestrated SRE reasoning loop in LangGraph with Silero VAD and memory checkpointer.
• Developed a real-time Next.js "War Room" dashboard streaming agent transitions via WebSockets.

Engineered a LangGraph agentic system for real-time microservice failure detection, correlation, and automated self-healing resolution actions.

• Cut manual intervention by 80% across active Kubernetes clusters.
• Integrated Prometheus & Loki for anomaly detection with Postgres audit logs for compliance.

Designed a microservices platform with continuous retraining on live market streams, automated drift detection, and zero-downtime container updates.

• Implemented LSTM and Transformer models to forecast market trends.
• Built robust automated alert triggers for MAE/RMSE score spikes.

Engineered an automated content repurposing pipeline using OpenAI Whisper for transcription and Gemini API RAG to generate platform-optimized posts.

• Processed 100% of input audio to transcripts without manual intervention.
• Reduced content creation time by 70% with customized multi-user workers.

VidXP is an open-source video indexing engine that connects LLM agents (Claude, Hermes, OpenClaw) to video understanding with low token overhead.

• Pull Request: #141: fix(fusion): bound moment components to shared overlapping interval (Closes #87)
• Core Search Fusion Fix: Resolved moment grouping bug in _connected_components / _shared_overlap_components where multi-modal search hits across video modalities produced sprawling intervals. Restructured fusion logic to tightly bound moments to their true shared overlapping window (max start & min end).
• Reliability & Test Coverage: Added extensive test suites in tests/test_search_fusion.py (+115 lines) covering temporal alignment, moment ordering, and boundary edge cases.