Projects
A tighter set of systems, ML, and robotics projects selected for technical depth, measurable outcomes, and implementation quality.
Input
Voice
Input
Notes
Input
Activity
GraphRAG index
Output
Memory
Output
Search
Output
Assist
Architecture
Local-first retrieval
Stack
Rust + Python + C#
A local-first personal OS that combines Rust, Python, and C# services with GraphRAG over Neo4j and ChromaDB to retrieve context from real user activity, notes, and voice transcripts.
Story 01
0.92
Story 02
0.81
Story 03
0.74
Ranking
MRR 1.0
Quality
NDCG@15 > 0.7
A graph-based news recommender built with heterogeneous user, article, and category graphs plus sentence-transformer embeddings, achieving MRR 1.0 and NDCG@15 above 0.7 while beating baseline methods.
Human
2%
AI
98%
Model
RoBERTa
Model
Llama
Model
Gemma
Corpus
1.2M+ texts
Accuracy
98%
A large-scale NLP pipeline that processed 1.2M+ texts and fine-tuned RoBERTa, Llama, and Gemma models to 98% accuracy for AI-generated content detection.
EN
99%
HI
98%
AR
97%
Dataset
300K+ samples
Accuracy
98-99%
A multilingual speech classifier trained on 300K+ audio samples using CNNs and audio transformers, reaching 98-99% accuracy across languages and accents.
Behavior
Real-time replanning
Core
D* + particle filters
A dynamic path-planning simulator that combines D* replanning and particle filters to navigate uncertain moving obstacles in real time.