Ahmad Kaleem Bhatti
AI ENGINEER
& BACKEND
ENGINEER
Building security infrastructure for autonomous AI agents. Founder of Actsurance. CS at Air University Islamabad.
Personnel File
I build backend systems and applied AI infrastructure. While completing a Computer Science degree at Air University Islamabad, I build independent products to solve specific operational problems. Python and Go handle the backend. TypeScript and Flutter handle the interfaces.
My current focus is Actsurance. Most production AI systems have no deterministic control layer between an autonomous agent and the external tools it can call. Actsurance intercepts every tool call, evaluates it against an Open Policy Agent (OPA) bundle, injects credentials through a Sealed Broker pattern, and either allows, denies, or escalates the request to a human reviewer. Every decision generates a cryptographically signed audit receipt.
Before Actsurance, I built Qualix AI to address lead leakage in messaging workflows. The system uses a RAG pipeline backed by LangChain and ChromaDB to retrieve relevant business context before qualifying a lead. The backend runs on FastAPI and PostgreSQL, with Celery and Redis handling the asynchronous processing that LLM calls require.
Outside of independent projects, I currently intern at Kaar-e-Kamal Welfare Foundation as an Application Developer Intern. I built the MVP of their Flutter and Go welfare platform, and now lead a team taking it to production. I also intern at Demetronics as a Full Stack Engineer Intern, where I recently migrated a decentralized IoT platform to a secure, centralized Express API. Everything points toward building production-grade, reliable systems as an AI Engineer and Backend Engineer.
I am currently completing my Bachelor of Science in Computer Science at Air University Islamabad (Expected Sep 2028). To further my technical foundations, I've completed certifications in Object-Oriented Data Structures in C++ (Coursera, UIUC), Fundamentals of Generative AI (Microsoft Learn), and CS50x (Harvard, edX).
Good engineering decisions are usually boring to describe. They involve choosing the option with fewer moving parts, writing things that are easy to change, and being honest about what you do not know yet.
Methods &
Tools
01
AI / LLM Systems
RAG pipelines, tool calling, agentic workflows
The Qualix AI RAG pipeline uses LangChain and ChromaDB to retrieve relevant business context before each inference call. The agent uses tool-calling endpoints to read and write CRM records without human intervention. Actsurance intercepts those tool calls and evaluates them against policy rules before execution.
02
Backend & APIs
Async APIs, data persistence, task queues
Qualix runs on FastAPI with async SQLAlchemy 2.0, PostgreSQL 16, PgBouncer, Redis 7, and Celery. The Kaar-e-Kamal welfare platform API is written in Go using pgx for database access and golang-jwt for token validation.
03
AI-Native Engineering
Designing workflows, orchestrating agents, rapid shipping
Built infrastructure and orchestrated AI agents to write, test, and ship code. Integrating tools like Cursor, Devin, and Antigravity into the development workflow accelerates the path from design to production by automating routine engineering tasks.
04
Security & Policy
Policy-as-code, audit logging, identity validation
Actsurance evaluates OPA policy bundles for deterministic RBAC decisions and scores prompt intent with an ONNX model for probabilistic routing. The gateway implements JWKS caching with a five-minute TTL, mTLS certificate rotation via cert-manager, and JWT ID reuse detection. These concepts were first tested in EMU86, a fail-closed 8086-assembly interrupt validator with hash-chained audit logs.
05
Frontend
Web dashboards, mobile applications
Qualix AI's web dashboards are built with React 19, Vite, and TypeScript. The public-facing site uses Next.js. The Kaar-e-Kamal welfare platform and PakLand property marketplace are both built in Flutter with go_router for navigation.
Work History
Building, learning & evolving...
Full Stack Engineer Intern
Demetronics (Private) Limited
A live IoT platform had exposed database credentials and JWT keys in the codebase, no centralized authorization model, and a mobile app that dropped usage history because it only wrote to local device storage.
Audited legacy systems to secure exposed secrets into environment variables. Architected a migration from decentralized client writes to a centralized Express API, implementing idempotent routes for device management. Introduced RBAC via Firebase Custom Claims. Redesigned the Flutter app's device service to provide offline local-IP fallback when the cloud API is unreachable.
Closed a live credential exposure across the platform, delivered a scalable Admin Portal for remote device management, and resolved the data-sync flaw that had been losing usage history on mobile.
Application Developer Intern
Kaar-e-Kamal Welfare Foundation
The foundation tracked Rashan distribution, Fee Assistance, Blood Donation matching, and Rozgar services through paper forms and manual records. Verification was slow and there was no role separation between field workers, operations staff, and administrators.
Solo-built the initial MVP of a Flutter mobile application and layered Go backend in one week. Now leading a team through the MVP-to-production phase to digitize all four welfare services. The system implements RBAC with JWT and Supabase Auth across four distinct roles with a multi-step verification flow.
Replacing a fully manual paper process with a structured, role-gated digital platform that enforces oversight at every step of the aid workflow.
Sales & Data Assistant
Atif E-Commerce Services
Inventory updates and customer communications across Amazon and eBay were handled manually. The volume of data made consistent accuracy difficult to maintain.
Wrote Python scripts using Pandas and Openpyxl to automate cross-platform inventory synchronization. Built a local desktop interface so non-technical staff could trigger the scripts without using a terminal. Added an LLM classifier to categorize incoming support emails before they reached the team.
Listing update time dropped by 35 percent. Support triage speed improved by 45 percent. Manual data correction work decreased by half.
Leadership
Guiding teams & building community...
Senior Lead Developer
AUCIS (Air University Computing and Innovation Society)
The computing society needed technical leadership to organize student engineering projects and bring junior developers up to production standards.
Joined as a developer and focused on shipping internal tools and mentoring newer members. Promoted to Senior Lead Developer in April 2026 based on these contributions.
Technical direction for all student-led engineering initiatives is managed under this role.
Management Team
Rizq
A social-impact program in Islamabad built around a 'Three Zero' theme. Intern teams had intense 10–12 day sprints to build real software solutions to local problems, featuring high-profile guest sessions from organizations like WWF-Pakistan and UNICEF.
Served on the management team handling program logistics (scheduling, venue, activities) and directly coordinated and mentored specific intern teams throughout their technical sprints.
Facilitated a structured environment where intern teams successfully shipped real solutions under strict deadlines.
Actsurance
CASE // ASSETActsurance
Challenge
AI agents execute tool calls against customer data, payment systems, and external APIs without any deterministic access control layer between them and the tools they invoke.
Investigation
Production AI systems often treat agent-to-tool communication as trusted. An agent that is tricked by a prompt injection attack, or that simply makes a bad inference, can execute a destructive tool call with no structural barrier to stop it. Actsurance sits between the agent and its tools. Every tool call passes through a policy evaluation layer before execution. The gateway produces a tamper-evident audit receipt for each decision, whether the call was allowed, denied, or escalated to a human reviewer.
Architecture
The L1 firewall uses the RE2 regex engine to inspect tool call payloads for SQL injection patterns, shell commands, and PII before any policy evaluation runs. A match returns HTTP 403 immediately.
Outcome
ESCALATE - Implementation starts with the L1 firewall. Selected to present at NIC Lahore and NIC Hyderabad.
Games Lab
Five playable simulations demonstrating adversarial search, heuristic evaluation, and constraint solving. Built entirely in React.
Chess Engine
StrategyPlay Chess against a computer that thinks multiple steps ahead to outsmart you.
- Play the computer
- Watch Auto-Play
- Adjust Difficulty
Connect 4
AdversarialDrop your chips and try to beat a computer programmed to predict and block your every move.
- Play the computer
- Watch Auto-Play
- Zero-Sum Game
2048
ProbabilisticWatch the computer automatically solve the 2048 puzzle by calculating the safest moves.
- Watch Auto-Play
- Grid Logic
- High Scores
Snake
PathfindingWatch the snake autonomously navigate the board to find food without crashing.
- Watch Auto-Play
- Real-Time Pathing
- Collision Avoidance
Minesweeper
LogicLet the computer safely flag mines for you using pure logic and probability maps.
- Play the computer
- Watch Auto-Play
- Safe Sweeping
Open to AI Engineering, Backend Engineering, and Agentic AI roles.