AGENTIC AI Β· AI SECURITY Β· RELIABILITY & RESILIENCE
I engineer AI systems that are secure, reliable, measurable, and resilient.
I build the engineering infrastructure around LLM applications and agentic AI workflows, with a focus on backend architecture, system behaviour, reliability, and security.
My work spans APIs, databases, retrieval pipelines, agent orchestration, testing, and evaluation.
I'm particularly interested in understanding:
- How AI systems fail and why.
- How vulnerabilities emerge in agent workflows.
- How to evaluate LLM and agent behaviour.
- How to engineer safeguards and recovery mechanisms.
- How to make AI applications more observable and dependable.
Current role: Full Stack Agentic AI Engineer @ Techloset
Professional direction:
AI Systems Engineering β AI Security β Reliability & Evaluation
|
|
|
|
- Controlled failure injection: timeouts, exceptions, and malformed responses.
- Structured execution traces and workflow state recording.
- Explicit invariant checks.
- Comparison of no-recovery, retry, and state-aware recovery policies.
- Reproducible experiment runner and persisted results.
Stack: Python Β· FastAPI Β· Pydantic Β· SQLite Β· Pytest Β· Docker
Evaluation includes a documented 45-run simulated baseline. This is a research prototype, not proof of production-wide agent reliability.
A reproducible collection of failure modes in RAG and agent workflows.
- Ungrounded answers and citation drift.
- Retrieval misses and context pollution.
- Tool loops and silent tool failures.
- Premature task termination.
- Numeric and date hallucinations.
- Reproduction, detection, and mitigation examples.
Stack: Python Β· Pytest Β· LLM Evaluation
A multi-agent RAG application with groundedness evaluation.
- Retrieval agent for relevant document context.
- Generation agent for grounded responses.
- Critic agent for evaluating generated answers.
- Citation and groundedness checks.
Stack: Python Β· FAISS Β· Sentence Transformers Β· Hugging Face Β· Gradio
A multi-tenant reporting application focused on backend integrity.
- Tenant-aware data separation.
- JWT authentication and role-based access.
- Validated CSV ingestion.
- PostgreSQL constraints and relational modelling.
- Approval workflows and audit records.
- Automated tests, migrations, and CI.
Stack: Python Β· FastAPI Β· PostgreSQL Β· SQLAlchemy Β· Alembic Β· Docker Β· Pytest
34 locally passing tests were recorded on 10 November 2026, including five PostgreSQL integration tests. Uses synthetic data.
Language-Conditioned Robotic Manipulation via Lightweight LLM Agents
Research direction:
- Natural-language task planning.
- Structured robotic action generation.
- Safety checks before execution.
- Simulation-based experimentation.
- Failure handling and self-correction.
Research interests: AI Safety Β· Robotics Β· Autonomous Systems
A pipeline for dataset validation and anomaly detection.
- CSV ingestion and validation.
- Data-quality analysis.
- Isolation Forest and Z-score anomaly detection.
- Interactive results and downloadable reports.
Stack: Python Β· Pandas Β· Scikit-learn Β· Gradio
An AI-powered career and resume analysis assistant.
- Resume-to-job matching.
- Skill-gap identification.
- ATS-style scoring.
- Resume optimisation.
- Cover letter and interview preparation.
- Application tracking.
Stack: Python Β· Streamlit Β· Groq API
An AI-assisted lead evaluation and prioritisation system.
- Multi-agent lead assessment.
- Historical deal context.
- Structured lead evaluation.
Recognition: IBM Granite Generative AI Hackathon winning team project.
| Organisation | Role | Period |
|---|---|---|
| Techloset | Full Stack Agentic AI Engineer | Nov 2026 β Present |
| MUSTER | Product Engineer | Jan 2026 β Oct 2026 |
| lablab.ai | Applied AI Developer | Aug 2024 β Aug 2026 |
| Stanford Code in Place | Section Leader | 2025 & 2026 |
| 1337 Ventures | Backend Developer | Jun β Sep 2025 |
| NIC Faisalabad | AI/Data Engineer | Jan β Aug 2024 |
- NIC Faisalabad: Developed Python ETL and reporting workflows supporting 30+ startups, associated with approximately 60% less manual analyst time.
- MUSTER: Contributed to an AI-powered real-estate lead qualification prototype.
- 1337 Ventures: Built backend components for an AI automation SaaS MVP.
- lablab.ai: Developed AI prototypes involving RAG, LLM integrations, and multi-step workflows.
- Stanford Code in Place: Supported Python learners through teaching, debugging, and problem-solving.
| Achievement | Recognition |
|---|---|
| IBM Granite Generative AI Hackathon | Winning Team β PolyEns |
| CS50 Puzzle Day | Team Cipher Cracker β 9 puzzles solved |
| Stanford Code in Place | Section Leader β 2025 & 2026 |
| BICON Accelerator, NUML | BillBachao selected for incubation |
| AlgoVerse Research Program | Accepted participant β 45% merit scholarship |
My current engineering focus is on developing deeper expertise in:
- LLM and agent failure modes.
- AI security and threat surfaces.
- Guardrails and safer tool execution.
- Reproducible testing and evaluation.
- Authentication and access control.
- Database integrity and PostgreSQL.
- Observability and distributed systems.
Bachelor of Science in Software Engineering
National University of Modern Languages (NUML)
2022 β 2026