flagship case studies documented in detail
Engineering Portfolio
I design AI products with systems-level rigor and production discipline.
My work spans end-to-end AI pipelines, LLM-powered applications, OCR and vision systems, eBPF-based telemetry, real-time voice bots, forecasting platforms, cloud infrastructure, and deployment tooling. I usually own the full path: architecture, training, evaluation, backend APIs, infra, rollout, and operational hardening.
AI Platforms
Kernel Observability
Voice Automation
Cloud Operations
additional projects across AI, infra, security, voice, and product engineering
core engineering tracks delivered end to end
Delivery style
Architecture to operations
I optimize for durable systems, not one-off demos.
Preferred scope
Multi-component products
APIs, workers, models, storage, observability, and rollout plans.
Operating principle
Pragmatic + measurable
Trade-offs are explicit, metrics are visible, and failure modes are handled.
Flagship Work
Selected projects that define how I build
These case studies show the kind of work I usually handle: multi-component architectures, production pipelines, model experimentation, traffic control, deployment, and real operational constraints.
AI Platform
Vision + OCR + LLMs
Infra
Haier AI Platform
Built multiple use cases under one program: retail store compliance, invoice extraction, legal document extraction, async processing, model routing, dashboarding, traffic shaping, Bedrock/Gemini integration, Redis-backed control plane, PostgreSQL persistence, and production monitoring.
- Designed and hardened inference pipelines and supporting infrastructure.
- Worked on model strategy, evaluation, invalid-image gates, and backend failover logic.
- Built admin and reporting capabilities for cost, throughput, usage, and operational visibility.
eBPF
Security Visibility
Kubernetes
eBPF Data Flow Agent, Nucleus & Backend
Designed a host-based outbound data-flow visibility platform to inspect server-side requests, classify traffic, study runtime-specific interception paths, move kernel events to user space efficiently, and relay captured data from agent to nucleus through Redis so the host-side agent stayed lightweight and decoupled from heavier processing.
- Explored ring-buffer and other kernel-to-user-space transport approaches.
- Worked on Python, Node.js, Java, and .NET interception directions for pre-encryption visibility.
- Used Redis as the practical transport layer between agent and nucleus for buffering, decoupling, and operational simplicity.
Voice Bot
FreeSWITCH
NLU + APIs
Air India Voice Bot Workflow
Built a voice workflow around FreeSWITCH, mod_audio_fork, ESL, separate playback control, NLU,
Air India API orchestration, and response-audio generation. The system streamed live call audio out, processed
intent externally, then played generated bot responses back into the call.
- Separated media outflow, bot intelligence, and ESL-driven playback return path.
- Implemented intent operations and workflow execution through API-backed voice automation.
- Created a practical call-control architecture rather than a simple demo bot loop.
Core Expertise
Where I usually contribute the most
AI systems & LLM workflows
I build production pipelines around LLMs, vision models, OCR systems, asynchronous job flows, validation layers, evaluation loops, prompt design, routing logic, and enterprise backend integration.
FastAPIBedrockGeminiQwen-VLClaudeRedis
Vision, OCR & model operations
I work on image quality gates, wrong-category detection, OCR extraction, model training pipelines, YOLO-based detection paths, PaddleOCR workflows, evaluation design, and structured output mapping for downstream systems.
YOLOv8PaddleOCRLayout-aware extractionEvaluation
Kernel, security & observability engineering
I explore eBPF-based capture, runtime interception, SSL/TLS boundary visibility, agent architecture, event transport, container-aware deployment, and backend intelligence around outbound data-flow analysis.
eBPFBCCByteBuddyASMLinuxKubernetes
Voice automation & conversational systems
I design telephony-to-bot workflows, audio streaming chains, speech handling, NLU orchestration, retry logic, TTS playback paths, and backend integration for real operational call flows.
FreeSWITCHESLWhisperAzure STTTTS
Cloud infra, APIs & platform hardening
I usually handle production runtime setup too: Nginx, workers, Redis queues, Postgres schemas, API auth, multi-worker tuning, S3 flows, Docker images, Kubernetes deployment, and admin/monitoring planes.
AWSAzureOCIDockerNginxPostgreSQL
Data platforms & forecasting systems
I also work on forecasting and analytical data products, building data contracts, feature layers, baseline models, causal features, and business-oriented forecasting workflows for real enterprise use cases.
ForecastingMySQLFeature pipelinesQuantile models
How I Work
My usual engineering pattern
Map the real problem
I usually start by narrowing the actual bottleneck: model accuracy, throughput, infra, latency, data quality, or runtime constraints.
Explore multiple approaches
I compare practical routes instead of forcing one tool. That may mean trying LLM-only flows, classical gates, runtime interception, or distributed queue designs.
Build the full pipeline
I usually own APIs, worker orchestration, model integration, storage, dashboards, load behavior, and deployment scripting together.
Harden for production
I care about failure modes, fallback logic, caching, rate limiting, monitoring, and making the system usable in real environments, not just in a demo.
Project Library
Broader work across AI, systems, platform engineering, and automation
Use the filters to focus on a track. The library combines flagship and additional projects so the website feels like a full profile, not just a small case-study list.
Downloads
Attached case studies
Haier AI Platform
PDF case study
Retail compliance, OCR workflows, legal extraction, infra, control plane, failover, monitoring.
eBPF Data Flow Agent & Nucleus PDF case studyKernel tracing, Redis transport to nucleus, deployment portability, container-aware observability, backend platform.
Air India Voice Bot Workflow PDF case studyFreeSWITCH, mod_audio_fork, ESL playback, NLU, API orchestration, generated response audio.