I build production-oriented AI systems that combine LLMs, agentic workflows, retrieval, machine learning, backend engineering, evaluation, and cloud-native infrastructure to solve practical engineering and business problems.

My focus is on building AI applications that go beyond prompt → response by incorporating structured outputs, deterministic validation, evidence, human-in-the-loop controls, testing, observability, and containerized deployment.

• Generative AI & LLM applications
• Agentic AI & workflow orchestration
• Retrieval-Augmented Generation (RAG)
• Document intelligence & structured extraction
• AI-assisted software engineering
• AI-powered QA & automation
• Machine learning & NLP systems
• Evidence-grounded AI systems
• Human-in-the-loop AI workflows
• Production-ready AI APIs and services
• Dockerized and cloud-deployable AI applications

An AI engineering platform that analyzes legacy repositories, builds a structured understanding of their architecture, creates migration plans, performs context-grounded code transformation, and evaluates the resulting migration.

Rather than sending an entire repository directly to an LLM, the system combines deterministic program analysis with agentic LLM workflows.

• Multi-language repository analysis
• Universal CTags symbol extraction
• Dependency and technology detection
• Complexity and structural signals
• Metadata-aware knowledge base
• LanceDB-backed retrieval
• Migration planning
• Context-grounded conversion agents
• Source/target structural comparison
• Migration risk scoring
• Unsupported-pattern detection
• Post-migration QA
• Migration reports and review checklists
• Provider abstraction for cloud/self-hosted LLMs
• FastAPI + Streamlit
• Docker Compose deployment
• Automated evaluation and CI

Legacy Repository
       ↓
Program Analysis
       ↓
Knowledge Base
       ↓
Migration Planning
       ↓
Agentic Conversion
       ↓
Post-Migration QA
       ↓
Risk / Gap Analysis
       ↓
Migration Report

Focus: Agentic AI · LLM Engineering · Code Intelligence · RAG · Software Modernization · FastAPI · Docker

A production-oriented AI automation platform demonstrating how LLM agents can become reliable asynchronous business workflows rather than isolated chatbot demos.

The platform supports content generation, competitor intelligence, partner outreach, and KPI/leadership reporting workflows.

• LangGraph workflow orchestration
• FastAPI control plane
• Pydantic contracts and validation
• Redis + RQ asynchronous workers
• PostgreSQL persistence and audit history
• Scheduled workflows
• Human-in-the-loop approval
• Evidence and validation
• External-impact action controls
• Deterministic demo mode
• Prometheus metrics
• Grafana dashboards
• OpenTelemetry tracing
• Jaeger distributed tracing
• CI evaluation suite
• Docker Compose deployment
• Worker scaling

Streamlit
    ↓
FastAPI Control Plane
    ↓
Redis Queue
    ↓
RQ Workers
    ↓
LangGraph Workflows
    ↓
Validation + Evidence
    ↓
PostgreSQL Audit Store

Observability
    ├── Prometheus
    ├── Grafana
    ├── OpenTelemetry
    └── Jaeger

Focus: Agentic AI · LangGraph · Distributed Systems · Async Workflows · Human-in-the-Loop · Observability · Production AI

A LangGraph-based research and document intelligence platform designed around a key principle:

An AI system should not claim evidence it cannot verify.

The system combines agentic routing, hybrid retrieval, reranking, web research, persistent conversation state, and deterministic citation verification.

• LangGraph agentic routing
• Knowledge-base / web / hybrid routing
• Dense + sparse hybrid retrieval
• Reciprocal Rank Fusion
• Qdrant vector search
• BM25 sparse retrieval
• Cross-encoder reranking
• Persistent conversation state
• Whole-document summarization
• Cross-document comparison
• Evidence registry
• Citation verification
• Citation coverage analysis
• Numeric/date evidence checks
• Evidence conflict detection
• Fail-closed behavior when evidence is insufficient
• Token instrumentation
• Evaluation suite
• Dockerized deployment
• CI

User Query
    ↓
LangGraph Router
    ├── Knowledge Base
    ├── Web Research
    ├── Both
    └── Direct Response
           ↓
    Hybrid Retrieval
           ↓
      RRF Fusion
           ↓
     Cross-Encoder
       Reranking
           ↓
      Evidence Set
           ↓
      LLM Synthesis
           ↓
    Citation Verification
           ↓
     Verified Answer

Focus: RAG · LangGraph · Hybrid Retrieval · Qdrant · Reranking · AI Reliability · Evidence Grounding

A production-oriented document intelligence system that processes multiple supplier quotations and turns unstructured commercial documents into structured, comparable and explainable procurement intelligence.

The system deliberately separates probabilistic LLM reasoning from deterministic business logic.

• Multi-document quotation ingestion
• PDF / DOCX / TXT / XLSX processing
• LLM-based structured extraction
• Provider-independent LLM architecture
• Gemini / OpenAI / optional Ollama
• Pydantic schema validation
• Generic output normalization
• Deterministic multi-criteria scoring
• Completeness validation
• Risk detection
• Price anomaly detection
• Evidence traceability
• Prompt-injection-aware document handling
• FastAPI API
• Streamlit dashboard
• Docker Compose
• Optional local/GPU inference
• Automated testing and CI

Supplier Documents
        ↓
Document Processing
        ↓
LLM Extraction
        ↓
Normalization
        ↓
Pydantic Validation
        ↓
Deterministic Scoring
        ↓
Risk / Anomaly Detection
        ↓
Evidence-Grounded Reasoning
        ↓
Recommendation
        ↓
Human Review

The core design principle is:

LLM → Extract & Reason

Python → Validate & Score

Human → Make the Final Decision

Focus: GenAI · Document Intelligence · Structured Outputs · LLM Reliability · Procurement Analytics · FastAPI · Docker

A Playwright-based web analysis and QA intelligence system that crawls dynamic websites, extracts structured DOM information, identifies interaction candidates, assesses QA risks, generates evidence-grounded test plans, and detects regressions.

The crawler is intentionally safe by default: it discovers and plans interactions but does not arbitrarily submit forms or trigger potentially destructive business actions.

• Playwright-based browser crawling
• Same-domain crawling
• Authentication support
• Dynamic-page analysis
• DOM structure extraction
• Heading and section analysis
• Form and field discovery
• Interactive-element detection
• QA risk scoring
• Evidence-grounded test generation
• Browser console-error detection
• Failed network request detection
• Regression baselines
• Structural change detection
• Optional Ollama summaries
• Markdown / HTML / JSON / CSV / DOCX reports
• Docker Compose
• Automated testing
• CI

URL
 ↓
Playwright Browser
 ↓
Same-Domain Crawl
 ↓
Structured DOM Analysis
 ↓
QA Risk Assessment
 ↓
Evidence-Grounded Test Generation
 ↓
Browser Health Signals
 ↓
Regression Comparison
 ↓
Reports

Focus: AI-Assisted QA · Playwright · LLM Applications · Test Intelligence · Browser Automation · Docker

Python · SQL

Machine Learning · NLP · Generative AI · Agentic AI · LLM Engineering

LangGraph · LangChain · CrewAI · Agno · OpenAI · Gemini · Ollama

Qdrant · LanceDB · Embeddings · Hybrid Search · BM25 · RRF · Cross-Encoder Reranking

FastAPI · REST APIs · Pydantic

Streamlit · Document Intelligence · AI Automation · QA Intelligence

Playwright

PostgreSQL · Redis · RQ

Prometheus · Grafana · OpenTelemetry · Jaeger

Docker · Docker Compose · GitHub Actions · CI/CD · AWS

I don't treat an LLM as the entire application.

I design AI systems as a combination of probabilistic intelligence and deterministic engineering controls.

             Unstructured Data
                    ↓
          Processing / Retrieval
                    ↓
              LLM / Agents
                    ↓
          Structured Validation
                    ↓
       Deterministic Business Logic
                    ↓
        Evidence / Quality Gates
                    ↓
        Evaluation & Observability
                    ↓
             API / Application
                    ↓
           Dockerized Deployment

This allows AI systems to become:

Reliable · Testable · Explainable · Observable · Deployable

Across my projects, I focus on:

• Structured LLM outputs
• Pydantic schema validation
• Deterministic scoring and business rules
• Evidence-grounded generation
• Citation verification
• Fail-closed AI behavior
• Human-in-the-loop workflows
• Prompt-injection awareness
• Automated evaluation
• Regression testing
• API health and readiness checks
• Asynchronous job processing
• Persistent audit trails
• Metrics and distributed tracing
• Containerized deployment

Building and deploying practical AI systems with a focus on:

Generative AI · Agentic AI · RAG · LLM Applications · AI Automation · Document Intelligence · AI Reliability · Production AI Engineering

LinkedIn: Connect with me

Live Projects:
See the pinned repositories above for deployed applications and technical documentation.

Portfolio:
Coming soon

Feel free to connect.