I build production-oriented AI systems that combine LLMs, agentic workflows, retrieval, deterministic controls, evaluation, observability, and cloud deployment.
My engineering focus is moving beyond prompt → response toward systems that are reliable, testable, evidence-grounded, observable, and deployable.
Multi-language legacy-code modernization platform combining program analysis, dependency intelligence, retrieval, migration planning, agentic transformation, and post-migration QA.
- AST / CTags analysis and dependency intelligence
- Qdrant-backed knowledge storage and retrieval
- Migration planning and context-grounded conversion
- Language-agnostic execution and validation adapters
- Post-migration quality gates and risk reporting
- FastAPI + Streamlit + Docker + Render Blueprint
Repository: https://github.com/AsaifAli/AI-Code-Modernization-Platform
Production-minded AI automation control plane built around asynchronous workflows, approval boundaries, evidence, persistence, scheduling, and observability.
- LangGraph orchestration
- FastAPI control plane
- Redis/RQ async execution
- PostgreSQL run and audit persistence
- Human-in-the-loop approval
- Prometheus, Grafana, OpenTelemetry, and Jaeger
- Render Blueprint deployment
Repository: https://github.com/AsaifAli/AI-Automation-Command-Center
LangGraph research and document intelligence system built around sparse-first, vector-optional retrieval, adaptive query recovery, evidence provenance, citation verification, and fail-closed behavior.
- OpenSearch inverted-index retrieval using BM25 for lexical relevance
- Exact, phrase, fuzzy, and metadata-aware sparse retrieval
- Neural-sparse retrieval without dense-vector k-NN, when enabled
- Dense-vector retrieval is not a hard dependency for the retrieval pipeline
- Reciprocal Rank Fusion (RRF) across independently ranked retrieval candidates
- Cross-encoder reranking with Jina AI
- Bounded agentic query expansion for weak first-pass retrieval
- Persistent multi-turn conversation state with LangGraph checkpointing
- Evidence registry with stable turn-scoped evidence IDs
- Whole-document evidence handling for stronger contextual synthesis
- Citation verification and claim grounding against the actual evidence set
- Numeric/date support checks and evidence-conflict detection
- Retrieved documents and web results treated as untrusted data, isolated from agent control flow
- Fail-closed behavior when evidence is missing, insufficient, unsafe, or cannot be verified
Repository: https://github.com/AsaifAli/LangGraph-RAG
AI-powered quotation analysis system that combines LLM extraction with deterministic validation and scoring.
- PDF / DOCX / TXT / XLSX ingestion
- Structured extraction and schema validation
- Deterministic procurement scoring
- Completeness and risk checks
- Evidence traceability
- FastAPI + Streamlit + Docker
Repository: https://github.com/AsaifAli/quotation-analyzer
Playwright-based web application QA intelligence platform for dynamic-site crawling, structured DOM analysis, QA risk assessment, test-plan generation, and regression detection.
- Same-domain browser crawling
- Authentication support
- Structured DOM and interaction analysis
- Evidence-grounded QA generation
- Browser health and regression signals
- Safe-by-default interaction model
Repository: https://github.com/AsaifAli/web-crawler-agent
A modular AI engineering platform for evaluating, observing, testing, and governing AI systems across RAG, agentic workflows, code modernization, document intelligence, and AI-assisted testing.
It currently integrates evaluation workflows for five AI systems:
- FlowPilot — Agentic automation control plane
- LegacyLens — Agentic software modernization
- EvidenceFlow — Verified sparse-first RAG
- QuoteSense — Procurement and document intelligence
- WebQA Intelligence — AI-assisted browser testing
| Capability | What it covers |
|---|---|
| AI Evaluation | Project-specific evaluation packs and golden datasets |
| RAG Evaluation | Recall@K, Precision@K, MRR, MAP@K, nDCG@K |
| Agent Evaluation | Task success, step success, tool-call validity, policy and approval compliance |
| Code Evaluation | Syntax validation, unit-test pass rate, semantic verification, release gates |
| Document Evaluation | Exact match, schema validity, numeric accuracy, risk F1 |
| Browser QA Evaluation | Test-generation P/R/F1, defect and regression detection, locator validity |
| LLM-as-a-Judge | Relevance, groundedness, faithfulness, completeness and qualitative scoring |
| Observability | Latency, tokens, cost, fallback and execution telemetry |
| Agent Tracing | Step-level execution traces and tool-call visibility |
| Guardrails | Prompt-injection detection, secret redaction and policy enforcement |
| Experiment Tracking | Baseline vs candidate comparison and regression detection |
| Prompt / Model Lineage | Versioned prompt, model, dataset and configuration tracking |
| Quality Gates | Automated thresholds for AI quality and regression control |
| Human-in-the-Loop | Approval workflows and auditable decisions |
| n8n Automation | Scheduled evaluation and regression workflows |
| Plugin Architecture | Reusable adapters and evaluation packs for new AI services |
The harness is designed as an extensible platform rather than a project-specific dashboard.
New AI services can be integrated through:
AI Service
↓
Project Adapter
↓
Normalized AgentRun
↓
Evaluation Pack
↓
Metrics + Quality Gates
↓
Observability / Audit
↓
Dashboard
Repository: github.com/AsaifAli/ai-engineering-evaluation-harness
The five interactive projects use a shared OpenAI-compatible gateway for BYOK sessions.
Portfolio
↓
Redis-backed session
↓
Short-lived JWT
↓
Project
↓
Portfolio LLM Gateway
↓
User's selected provider/model
Provider credentials remain server-side; projects receive a temporary session token rather than the provider API key.
Repository: https://github.com/AsaifAli/shared-Portfolio-LLM-Gateway
Unstructured Data
↓
Processing / Retrieval
↓
LLM / Agents
↓
Structured Validation
↓
Deterministic Logic
↓
Evidence / Quality Gates
↓
Evaluation / Observability
↓
API / Application
↓
Containerized Deployment
Across projects I emphasize:
- Structured LLM outputs and schema validation
- Evidence-grounded generation and citation verification
- Retrieval designed for recall, precision, and transparent failure modes
- Deterministic business rules and quality gates
- Human-in-the-loop controls for external-impact actions
- Automated evaluation and regression testing
- Health/readiness checks and observability
- Dockerized, cloud-deployable services
LinkedIn: https://www.linkedin.com/in/sk-asaif-ali-134873243/
Portfolio: https://asaifali-portfolio.vercel.app
GitHub: https://github.com/AsaifAli
Generative AI · Agentic AI · RAG · LLM Systems · AI Reliability · AI Automation · Document Intelligence · Software Modernization · AI-Assisted QA