Open to Full-Stack & Forward-Deployed AI Engineer Roles

Hamza AliFull-Stack AI & LLM Systems Engineer — Multi-Agent Workflows, Enterprise RAG & MCP

Production AI systems that prove their answers.

I build the whole stack so production AI staysaccess-controlled.

Full-stack AI engineer building the whole vertical slice: FastAPI backends, React frontends, hybrid retrieval pipelines, MCP tool servers, eval harnesses, and Docker Compose deployments — ready to partner directly with product and client teams to ship AI into real workflows.

Lahore, Pakistan · Open to Remote (UTC+5)FastAPI · React · Evals · Docker Compose
690+ Tests
AI Code Review CI matrix (Py 3.10–3.13) + 14-case regression
90 Tests · 0.93
FinSight API, 3-layer RBAC (0 filter leaks) & OWASP suites
< 20ms
Role-scoped Redis replay (0.84 / 0.894 answer relevancy)
~1M Rows
NOAA lightning telemetry & Tableau risk map at ATS AI Lab

[01//flagship work]

Three end-to-end AI platforms, built for real failure modes.

Packaged with Docker Compose and documented with Architecture Decision Records, evaluation harnesses, and narrated walkthroughs — inspect the full code and trade-offs behind each system.

AI Code Review Agent walkthrough preview
01Flagship · Multi-Agent PR Review

Docker Compose + Caddy TLS · GitHub Action · Stdio MCP

Zero-token static/AST pre-scans + 3-agent CrewAI Flows crew + plain-code deterministic reconciler, durable SQLite WAL queue, 8 MCP tools, and 690+ automated tests.

690+Automated Tests (Py 3.10–3.13)
100.0%14-Case Regression Suite
8 MCP · 15 CmdsIDE & PR Bot Surface
Python 3.10–3.13CrewAI FlowsFastAPITree-SitterSemgrep / Bandit / Ruff
FinSight walkthrough preview
02Enterprise RAG · 3-Layer RBAC

Docker Compose Full Stack (FastAPI + React + Redis)

3-layer RBAC (0.93 security score, 0 filter leaks), 3-way query expansion + ChromaDB/FTS5/Cohere RRF, sandboxed in-memory DuckDB SQL, <20ms Redis cache, and 90 automated tests.

90Automated Tests Passed
0.93RBAC Security Score
PythonFastAPIReact 19
Factscope AI walkthrough preview
03Web + Desktop · Live Search Grounding

Web Serverless + Electron 36 Desktop

Deconstruct articles or URLs into atomic claims, verify against live Google Search, and degrade transparently across a 3-tier Gemini failover engine.

3 TiersAutomatic AI Failover
24hNormalized Result Cache
Gemini 2.5 FlashGoogle Search GroundingReact 19

[02//systems lab]

Inspect the architecture, protocols, evals & token economics.

Step through authorization boundaries and fallback states, inspect all 8 MCP JSON-RPC tool schemas, review scoped repository benchmarks, and model LLM caching costs.

Illustrative example

PR diff adds hallucinated package `flask-auth-helpers` + `subprocess.run(shell=True, input=user_input)`.

Step 1 / 4

Pre-Ingestion Shield

Strip Unicode Bidi characters, validate HMAC/Bearer auth, and scan dependency manifests.

flask-auth-helpers -> SEC-SLOPSQUAT-001 (hallucinated dependency flagged)

[03//story & experience]

Owning the whole stack from retrieval to deployment.

The Engineer Behind the Systems

Hamza Ali

Full-Stack AI

Full-stack engineer focused on AI systems that hold up outside a notebook. I build the entire vertical slice — FastAPI backends, React frontends, hybrid retrieval pipelines, evaluation harnesses, and Docker Compose deployments — so nothing gets lost in handoff between prototype and production.

Most AI demos skip the unglamorous engineering that actually decides whether a team can trust the system: per-request access control before the LLM ever runs, deterministic static scanners that constrain what agents can claim, durable webhook job queues, and automated regression tests (90 in FinSight, 690+ in AI Code Review Agent) that measure when the model is wrong.

Before focusing full-time on LLM systems, I bridged academic CS foundations and operational analytics at ATS AI Lab — turning ~1 million NOAA lightning-strike records into Tableau risk-zone recommendations — and regularly mentored fellow engineers on architecture and technical decisions. I'm looking for a full-stack or forward-deployed AI engineering role working directly with product and client teams to get AI into real workflows.

Engineering Standard

Every system ships with clear access boundaries, reproducible Docker Compose packaging, automated regression tests (90 & 690+), and honest measurement of where models fail.

Independent AI / Full-Stack Engineer

Self-Directed & Open-Source Production Systems

Nov 2024 – PresentLahore, Pakistan · Remote

End-to-end engineering of full-stack LLM platforms — owning FastAPI backends, React 19 TypeScript frontends, retrieval and agent orchestration, access control, automated testing (780+ tests), and Docker Compose deployment.

  • Shipped FinSight — enterprise RAG + Text-to-SQL platform with 3-layer RBAC (0.93 security score, zero metadata-filter violations), hybrid ChromaDB + FTS5 + Cohere retrieval, DuckDB SQL sandbox, sub-20ms Redis cache replay, and 90 automated tests.
  • Built AI Code Review Agent — deterministic-first multi-agent PR review platform (3-agent CrewAI Flows crew + plain-code reconciler), durable SQLite WAL webhook queue, 15 PR slash commands, 8-tool MCP server, SARIF v2.1.0, and 690+ tests across Python 3.10–3.13.
  • Delivered Factscope AI — full-stack claim decomposition & Google Search-grounded verification platform with 3-tier quota failover across Web and Electron 36 desktop.
PythonFastAPIReact 19TypeScriptCrewAI FlowsMCPDocker ComposeRAGAS

Data Science Intern

Advanced Telecom Services (ATS) AI Lab

Aug 2024 – Oct 2024Microsoft Azure ML Program

Turned large-scale raw geospatial weather telemetry into stakeholder-ready operational risk intelligence alongside Microsoft Azure ML cloud workflows.

  • Ran end-to-end EDA and feature engineering on ~1 million NOAA lightning-strike telemetry records (2009–2018), alongside Microsoft Azure ML training.
  • Built an interactive Tableau strike-density heatmap of US lightning activity, identifying Texas and the southeastern US as the highest-risk zones.
  • Presented risk-zone analysis and operational mitigation recommendations directly to non-technical stakeholders.
PythonPandasTableauMicrosoft Azure MLETLEDA

BS Computer Science

Abbottabad University of Science and Technology

2022 – Jul 2026Pakistan

Computer Science fundamentals applied directly to distributed AI systems, access-controlled data stores, and peer technical mentorship.

  • Final Year Project: FinSight AI — An Enterprise RAG & Text-to-SQL Platform with Role-Based Access Control.
  • Peer teaching: regularly taught and guided fellow students on technical skills, architecture trade-offs, and project decisions.
Computer SciencePeer TeachingFYP: FinSight AI
Apr 2026

DeepLearning.AI

Multi-Agent Systems with CrewAI

Role-scoped agent crews, provenance guardrails & tool orchestration

Dec 2025

DeepLearning.AI

Retrieval Augmented Generation

Chunking, hybrid dense+sparse retrieval, reranking & RAG evals

Nov 2025

DeepLearning.AI

Deep Learning Specialization

Neural network architectures, optimization, sequence models & diagnostics

Oct 2024

Stanford Online

Machine Learning Specialization

Supervised & unsupervised learning, bias/variance & model evaluation

[04//engineering notes]

How I think about failure modes & trade-offs.

Short architectural notes written directly from production trade-offs encountered while building FinSight, AI Code Review Agent, and Factscope AI.

Note 01FinSight3 min note

Why Authorization Must Happen Before Retrieval & LLM Routing

Prompt instructions are not an access-control boundary. In enterprise RAG, permissions must be enforced before any embedding search, SQL query, or model call.

Many RAG prototypes retrieve top-k chunks first and tell the LLM in the system prompt to 'ignore documents the user should not see.' That pattern fails immediately under prompt injection and still leaks sensitive tokens into model context.

In FinSight, department access runs across three independent layers: JWT verification with per-request DB role re-validation, a zero-token pre-LLM department guard, and hard ChromaDB metadata filters plus an isolated in-memory DuckDB table whitelist (`enable_external_access=False`).

The automated RBAC security suite scored 0.93 with zero metadata-filter violations, while transparently surfacing a general-doc 'revenue' phrase calibration note — proving why access boundaries must be measured with regression tests, not assumed.

Key Engineering Takeaways

  • Re-validate the user's current database role on every JWT request so revoked roles cannot query stale permissions.
  • Block cross-department intent in a pre-LLM guard to eliminate both data leakage and wasted token spend.
  • Isolate Redis cache keys by role digest (`finsight:query:{role}:{digest}`) so repeat queries replay in < 20ms without cross-tenant bleeding.
Inspect full FinSight case study
Note 02AI Code Review Agent4 min note

Run Deterministic Checks First to Limit What Agents Can Claim

Sending raw diffs straight to an LLM invites hallucinated reviews. Running static scanners, AST call graphs, and a plain-code reconciler constrains agent output to verifiable facts.

Note 03Factscope AI3 min note

Designing LLM Backends That Degrade Honestly Under Quota Pressure

LLM APIs and search grounding tools have separate rate limits. Production apps need multi-tier failover, error classification, and transparent UI state when grounding is unavailable.

[05//verified stack]

Tools proven directly inside shipped repositories.

Every framework, database, and security layer listed below maps to real code in FinSight, AI Code Review Agent, Factscope AI, or ATS AI Lab — zero resume filler.

Google Gemini & GenAI SDK

Grounded Generation, Routing & Failover

Gemini 2.5 Flash & 3.1 Flash Lite with Google Search grounding, structured outputs, fallback chains, and gemini-embedding-2-preview.

CrewAI & LangChain

Bounded Multi-Agent Crews & LCEL

Multi-agent review crews with bounded tool loops, opt-in cross-model adversarial critique, and LCEL RAG retrieval pipelines.

Model Context Protocol (MCP)

8-Tool Stdio MCP Server

Exposes SAST scans, governance rules, Tree-Sitter impacted callers, semantic context, team memory, and test generation to Claude Code & Cursor.

RAGAS, DeepEval & Deterministic Evals

Dual-Track Quality & Security Harnesses

LLM-as-judge (RAGAS Answer Relevancy 0.894, Faithfulness 0.850; DeepEval G-Eval) paired with quota-free cosine/BM25/ROUGE-L & regression gates.

ChromaDB + SQLite FTS5 (RRF)

Dense + BM25 Reciprocal Rank Fusion

Role-partitioned ChromaDB cosine search fused with SQLite WAL/FTS5 BM25 keyword retrieval using Reciprocal Rank Fusion (k=60).

Cohere Rerank v3

Cross-Encoder Context Reranking

Optional second-stage cross-encoder scoring (rerank-english-v3.0) promoting the top-6 highest-signal passages with graceful RRF fallback.

DuckDB In-Memory Sandbox

Isolated NL-to-SQL Execution

In-memory DuckDB sandbox (`enable_external_access=False`, 256MB cap, 10s interrupt timer, 1,000-row limit) with automatic SQL -> RAG fallback.

Redis 7 Role-Scoped Cache

Sub-20ms Streaming Replay

Multi-tier Redis 7 + in-memory cache using SHA-256 role-isolated keys (`finsight:query:{role}:{digest}`), 300s TTL, and upload invalidation.

Python 3.10–3.13 & FastAPI

Async Streaming APIs & Workers

Typed FastAPI + Uvicorn services with Pydantic validation, SSE & NDJSON token streaming, durable SQLite job queues, and Pytest harnesses.

Tree-Sitter Polyglot AST

6-Language Call & Taint Graphs

Cross-file symbol resolution and impacted-caller blast radius across Python, TypeScript, JavaScript, Go, Java, and Rust.

Semgrep, Bandit, Ruff & SARIF

Deterministic Security & CI Gates

Layered static analysis, Trojan Source / slopsquatting shields, OASIS SARIF 2.1.0 export, and zero-cost `.code-review.yaml` governance.

JWT RBAC, SlowAPI & OWASP Guards

Defense-in-Depth Authorization

HS256 JWT + live DB role re-fetch, SlowAPI tiered limits (/login 5/min, /upload-docs 10/min, /chat 30/min), and 20 OWASP security tests.

React 19, TypeScript & Tailwind v4

Streaming Workbenches & SPAs

Live annotated code viewers, interactive SVG blast-radius graphs, progressive NDJSON chat interfaces, and Electron 36 desktop packaging.

Docker Compose & Hardened Containers

Reproducible Self-Hosted Stacks

Multi-stage Docker builds with non-root UID 10001, read-only root FS, `cap_drop: ALL`, Caddy automatic TLS, and persistent volumes.

GitHub Actions, CI Gates & Webhooks

Reusable Action & Signed Gates

Reusable GitHub Action, Python 3.10–3.13 CI matrix, mypy, pip-audit, Gitleaks secret scanning, Playwright E2E, and PR slash commands.

Pandas, Tableau & Microsoft Azure ML

Telemetry Pipelines & Risk Analytics

Processed ~1M NOAA lightning-strike telemetry records into Tableau risk heatmaps and operational recommendations at ATS AI Lab.

[06//collaboration]

Open for full-stack & forward-deployed AI roles & selective advisory.

Looking for a full-stack or forward-deployed AI engineering role where I work directly with product and client teams to get AI into real workflows — plus selective architecture & evaluation engagements.

Primary Focus · Full-Time & Forward-Deployed Roles

Hiring a Full-Stack or Forward-Deployed AI Engineer?

I own the full vertical slice — FastAPI backends, React/TypeScript UIs, hybrid RAG + 3-layer RBAC, CrewAI/MCP agent workflows, 780+ automated tests, and Docker Compose deployments — working directly with engineering and client teams remotely (UTC+5).

Advisory 01

Enterprise RAG & Access-Controlled Search

Production knowledge workspaces enforcing department permissions before retrieval, blending dense + BM25 hybrid search, and isolating structured SQL analytics.

  • 3-layer RBAC: JWT role re-validation, pre-LLM guard & store filters
  • Hybrid retrieval (ChromaDB + BM25 + RRF + Cohere reranking)
  • Sandboxed in-memory DuckDB Text-to-SQL with automatic RAG fallback
  • Role-scoped Redis cache (< 20ms replay) & OWASP regression suite

Deliverable: Architecture + Tested Implementation + Docker Compose Handoff

Discuss this engagement
Advisory 02

Deterministic AI Agents & MCP Tool Servers

Multi-agent workflows and stdio MCP servers grounded in static/AST pre-scans, plain-code reconciliation, durable webhook queues, and fail-closed policies.

  • Zero-token static & AST pre-scans constraining what agents claim
  • CrewAI Flows / LangGraph crews with plain-code deterministic scoring
  • Custom MCP tool servers for Claude Code, Cursor & internal IDEs
  • Durable FastAPI webhook queues, slash commands & SARIF v2.1.0

Deliverable: Agent Pipeline + MCP Server + CI/SARIF Integration

Discuss this engagement
Advisory 03

LLM Reliability Audits & Evaluation Harnesses

Measure when models are wrong and harden brittle prototypes with RAGAS/DeepEval harnesses, quota-free regression suites, failover chains, and cost/latency tuning.

  • Dual-track evals (RAGAS, DeepEval & quota-free exact-match gates)
  • Multi-tier model & search failover with exponential backoff retries
  • Role-scoped caching, rate limiting & token-cost profiling
  • Prompt-injection hardening & OWASP LLM security regression suites

Deliverable: Technical Audit + Automated Eval Harness + Remediation Plan

Discuss this engagement

[07//contact]

Let’s talk about your team or AI system.

Direct email response · Remote (UTC+5)

Role or project inquiry

Share the role, problem, or system goal.

Whether you’re hiring a GenAI engineer or scoping a RAG/agent system, a few lines of context helps me respond meaningfully.

Add role or project details optional

Your details are used only to respond to this message. Automated spam screening protects this form.

direct contact
Lahore · UTC+5

Hamza Ali

Full-Stack AI & LLM Systems Engineer · Python, FastAPI, React & LLM Systems

Open to full-time remote Full-Stack AI and Forward-Deployed AI Engineer roles working directly with product or client teams, as well as selective consulting on RAG, multi-agent pipelines, and evaluation harnesses.

Open to Full-Stack & Forward-Deployed AI Engineer Roles
Download Resume / CV (PDF)