MODEL LAYER
Foundation reasoning & multimodal intelligence powering structured workflows.
GPT-4oClaude 3GeminiLlamaHuggingFace
Used in multi-agent orchestration, structured RAG, and enterprise automation.
1.5+
Years of hands-on experience
Tech Stack
Vector DBs & RAG
AWS S3 (boto3)DockerDockploy
Storage
SQLAlchemy 2AlembicUUID / JSONB
Backend & Auth
UvicornPydantic / pydantic-settingshttpx (async client)Argon2 (argon2-cffi)
"Life is not a gallery of flawless victories — it's a Museum of Failures and a Gallery of Trying."
— Annie Siri
Intelligence Architecture
API & INTEGRATION LAYER
Provider abstraction and multi-model switching via unified interfaces.
OpenAI APIAnthropic APIGemini API
Client → API Gateway → Model Provider
DATA & RETRIEVAL LAYER
Contextual grounding using vector search and structured retrieval pipelines.
AWS S3 (boto3)DockerDockploySQLAlchemy 2AlembicUUID / JSONB
User Query→Embed→Vector Search→Context Injection→Model
AGENT ORCHESTRATION
Stateful, multi-step reasoning systems with graph-based execution.
LangGraphCustom Agent Systems
Planner→Tools→Memory→Executor→Output
BACKEND & SECURITY
Authenticated, access-controlled, production-grade APIs.
UvicornPydantic / pydantic-settingshttpx (async client)Argon2 (argon2-cffi)Bearer AuthenticationEnv-based SecretsReact Router 7Vite 6Tailwind CSS 4Framer MotionLucide ReactTypeScript
RBACRate LimitingSecure Token Handling
PERCEPTION & DEEP LEARNING
Vision pipelines and domain-specific model training.
Deep LearningPyMuPDF (PDF reading / thumbnails)Pillow (image processing)pypdf
CNNMultimodal ProcessingEdge Inference
DEPLOYMENT & MONITORING
Scalable deployment with logging, monitoring and CI/CD.
PyMuPDF (PDF reading / thumbnails)Pillow (image processing)pypdf
Latency · Error Rate · Throughput · Cost
CLOUD INFRASTRUCTURE
Elastic compute across cloud and serverless environments.
AWSDockerDokploy
Selected Projects
Kutum – The Family Information OS
Problem
Families juggle scattered data points—documents, health records, milestones—across chats and…
Solution
Modular architecture centered on three domains: People, Documents, and Health.
Impact
No more remembering dates or digging through WhatsApp.; Documents renewed on time; health…
Click to view details and links
More projects
Player Re-Identification System
YOLOv8
Developed a computer vision system using YOLOv8 and ByteTrack for player detection and tracking in sports videos, ensuring consistent ID assignment across frames. Implemented with Python, OpenCV, and boxmot to process 15-second 720p video, handling player re-entry with robust multi-object tracking. Documented with detailed README and shared on GitHub.
Tech
YOLOv8ByteTrackOpenCV
Deep Dive Case Studies
Narratives of engineering journeys, from architectural decisions to deployment challenges.
Recent Blogs
View AllNo blog posts yet.