Professional Portfolio
I build AI products, observability systems, voice platforms, and production infrastructure that actually run at scale.
I work across AI/ML pipelines, OCR and vision systems, speech automation, kernel-level observability, LLM security, infrastructure setup, throughput optimization, and deep troubleshooting. My strongest work sits at the intersection of product engineering, systems design, and hard technical execution.
AI / ML Platforms
Voice & Speech Systems
eBPF & Observability
Security Engineering
Infra & Kubernetes
Performance Optimization
Flagship Work
Most important projects first
These are the projects that best represent my depth: end-to-end architecture, implementation, infrastructure, production tuning, experimentation, and execution under real constraints.
01 · Enterprise AI Platform
Haier AI Platform
I built major parts of Haier’s AI program end-to-end: retail store compliance, invoice extraction, legal document extraction, shared async platform services, evaluation flows, and the surrounding production infrastructure.
Retail store compliance pipelines
OCR + LLM JSON extraction
Legal document summarization/extraction
FastAPI + Redis + PostgreSQL + S3 patterns
Gemini / Bedrock routing and failover
Admin dashboard, cost, traffic, and latency monitoring
Rate limiting, inflight control, burst shaping
Infra isolation, Nginx routing, worker tuning
02 · Kernel / Security / Platform
eBPF Data Flow Agent, Nucleus & Backend
I worked on a host-based outbound data-flow visibility platform that traced what servers were sending out, explored kernel-to-user-space transport choices such as ring buffer, and evolved into an agent → nucleus → backend architecture.
Outbound HTTPS/API visibility focus
Ring buffer and kernel/user transport exploration
Redis transport from agent to nucleus
Python, Node.js, Java, .NET interception paths
Java SSL/TLS instrumentation via ByteBuddy & ASM
Dynamic kernel-aware compilation/preparation
Hybrid containerized agent with host-kernel access
Kubernetes DaemonSet deployment model
Redis was chosen as the practical relay from agent to nucleus to keep the host-side agent light, decouple capture from analysis, buffer bursts, and make downstream enrichment/processing easier to control.
03 · Voice Automation
Air India Voice Bot Workflow
I built a voice-bot workflow around FreeSWITCH, mod_audio_fork, ESL, NLU, API orchestration,
and response audio injection. The system streamed audio out, processed intent and workflow logic remotely,
generated response audio, and played it back into live calls.
FreeSWITCH telephony flow
mod_audio_fork outbound audio stream
ESL-driven playback control
Bot-side intent handling and API calls
Air India workflow orchestration
Generated response audio + playback script
Retry/welcome/replay flow thinking
Telephony + NLU + TTS integration
Major Platforms & High-Value Work
Additional projects that matter
These projects expand the profile beyond the three flagship case studies and show the broader range of production, data, security, and engineering work I have handled.
Hindware Voice Intelligence Pipeline
Converted customer-care voice data into text at very large scale, then used LLaMA-driven extraction and downstream processing to turn unstructured voice interactions into structured database-ready information.
- 20 lakh+ audio files processed into text-oriented records
- Data cleaning, normalization, and transcript handling
- Pincode generation, city mapping, NER and parameter extraction
- Pipeline design for large-scale structured ingestion
Observa AI · Agent Observability
Built observability thinking and architecture around AI agents to track what they are doing, where they are sending data, and how their actions can be audited or monitored in a more reliable way.
- Agent activity and behavior tracing
- Data destination awareness
- Action visibility and monitoring concepts
- Security and governance-oriented architecture
7-Layer LLM Security System
Designed a layered defense system where the LLM acted as the final layer rather than the only layer, focused on detecting malicious prompts and preventing attempts to extract protected data through prompt attacks.
- Seven-layer defensive model
- Prompt abuse and data-extraction detection
- Layered validation before model escalation
- Security-first architecture for LLM usage
Sales Forecasting, DataCube & DataQube
Worked on data and forecasting initiatives involving structured data pipelines, baseline forecasting logic, feature contracts, and business-facing analytical workflows.
- Forecasting pipeline design
- Feature-layer and data-contract thinking
- Business-aligned analytical architecture
- Integration-ready structured data planning
Vision, OCR & Evaluation Work
Designed and refined image and document intelligence pipelines involving OCR, layout-aware extraction, evaluation logic, wrong-category detection, quality gates, and structured output generation.
- YOLO / category detection strategy
- PaddleOCR-style extraction direction
- Quality-gate thinking for invalid inputs
- Evaluation, schema, and reliability design
Throughput Optimization & Production Hardening
I regularly optimize web applications and services by improving request throughput, reducing bottlenecks, tuning workers, fixing database queries, and implementing caching and control-plane strategies.
- Database query optimization
- Redis caching and request shaping
- Latency and throughput tuning
- Worker/runtime and infra-level hardening
Cross-Stack Troubleshooting & Technical Support
A recurring part of my work is unblocking developers when they get stuck anywhere across the stack — from Linux shell issues and Docker problems to system configuration, deployment, driver issues, and runtime debugging.
- Linux, Docker and system troubleshooting
- Nginx, deployment, CI/CD and infra debugging
- Database, Redis and API integration support
- Fast technical problem-solving across domains
Architecture Mindset
I usually own the whole path, not just one layer
My strongest work is end-to-end: product problem understanding, architecture choice, pipelines, infra, monitoring, security, optimization, deployment, and practical fixes when production reality doesn’t match theory.
AI pipeline ownershipTraining, evaluation, OCR, voice, extraction, routing, fallbacks
Systems depthLinux, kernel thinking, containers, DaemonSets, runtime instrumentation
Operational maturityDashboards, observability, admin tools, logs, cost and throughput awareness
Capabilities
Where I bring the most value
My profile is best described as AI/ML plus systems plus platform plus security plus infrastructure — not a narrow single-role track.
AI & ML Systems
LLM workflows, OCR extraction, vision systems, structured output design, evaluation logic, and applied ML pipelines.
Voice & Speech Platforms
Speech-to-text pipelines, telephony integration, bot workflows, NLU orchestration, audio response loops, and call automation.
Kernel & Observability
eBPF concepts, transport from kernel to user space, runtime interception, outbound data-flow analysis, and host observability design.
Infrastructure & Deployment
Docker, Kubernetes, Nginx, FastAPI, Redis, PostgreSQL, system services, cloud deployment, and environment hardening.
Performance & Reliability
Throughput tuning, database optimization, cache strategy, concurrency shaping, fallback patterns, monitoring, and operations.
Security-Oriented Engineering
Prompt security, payload sensitivity classification, outbound data-awareness, AI-agent observability, and layered defense design.
Documents
Portfolio case studies
Detailed PDF write-ups for the three most developed case studies currently packaged with this site.
Featured PDF
Haier AI Platform
Retail compliance, OCR, legal extraction, infra, routing, failover, and admin-plane monitoring.
Open PDF → Featured PDFeBPF Agent & Nucleus
Kernel/user-space design, Redis relay to nucleus, runtime interception, and hybrid container deployment.
Open PDF → Featured PDFAir India Voice Bot
FreeSWITCH, mod_audio_fork, ESL, NLU, API workflows, and response-audio playback path.
Open PDF →