Building softwarethat builds software.

Workflow Automation SaaS — engineering workflows, not pages

A form submission event triggers a workflow engine that orchestrates an AI agent, routing, conditions and integrations like Slack, Discord and email, persisting results to PostgreSQL for analytics. Long-running AI steps execute asynchronously on background workers with Inngest, so concurrent workflows never block each other. Failed steps retry with backoff until they succeed. The system scales to dozens of concurrent workflows with usage-based billing via Polar.

OpenRouter — one API, infinite intelligence

A unified AI gateway: one prompt enters, and the router evaluates latency, cost, availability and context window across Claude, GPT, Gemini, DeepSeek and Mistral before selecting a provider. Whichever model answers, exactly one response returns through the same API — provider independence as infrastructure. Usage and cost are tracked per provider, and the gateway calmly routes a hundred concurrent requests. Built with Elysia on Bun, JWT auth and Prisma.

Polaris — the IDE that thinks before it types

An agentic AI editor. Given the prompt “Build OAuth authentication,” the agent pauses to reason before acting: it plans, retrieves documentation through Firecrawl, and opens only the relevant project files for context. Three workers — architecture, implementation and testing — run in parallel and merge. The agent writes real authentication middleware, hits a TypeScript error during the build, reads the error, fixes the faulty line, and re-runs until the build passes and the preview refreshes. Built with CodeMirror, Convex, Firecrawl and Clerk. When the work is done, the editor collapses until only a terminal remains.

Vertex — a coding agent living entirely inside the terminal

An autonomous terminal coding agent. Given “build rate limiter,” it thinks, plans and reads the repository before acting; it selects tools deliberately — search, filesystem, editor, shell — writes a sliding-window rate limiter, and runs the tests. One test fails: the bucket key rolls over mid-request. The agent reads the stack trace, clamps the window timestamp, re-runs the suite and lands three passing tests. Autonomy, not autocomplete: it recovers from failure without being asked. Its final command, “explain MCP,” becomes the next chapter's search query.

Poorplexity — search is retrieval, intelligence is synthesis

An AI answer engine. The query “explain MCP” fans out across the web; Firecrawl retrieves six pages and strips them to three clean sources. The extracted knowledge streams into a reasoning core that synthesizes a concise answer, with citations landing one after another, each traceable back to its source. Retrieval is the easy part — synthesis is the intelligence.

BOCK — intelligence across modalities

A unified multimodal pipeline, built during an AI/ML internship at BOCK Health. One video enters and branches into vision, audio, text and video lanes; each modality is processed independently while cross-modal attention lets the lanes inform each other. The four streams converge in a fusion engine that produces a single unified output. Afterwards, every system in this portfolio pulls back into one constellation — all of them built by one engineer.

The six systems just explored — workflow automation, the OpenRouter gateway, the Polaris editor, the Vertex terminal agent, the Poorplexity answer engine and the BOCK multimodal pipeline — settle into one constellation. Each keeps a faint habit from its chapter. They were all engineered by one person.

Experience and technical foundations

Experience as engineering responsibility. At BOCK Health, an AI internship building three foundational models unified into the multimodal pipeline shown earlier. At JK Tyres, data science — analytics and statistics over live manufacturing data. At SS Technology, computer vision for PCB manufacturing with automated defect detection. Technical foundations span TypeScript, Python and SQL; transformers implemented from first principles, multimodal fusion, retrieval and MCP; PostgreSQL, Prisma, Redis and event-driven jobs; orchestration, retries, routing gateways and rate limiting; agent runtimes and editor tooling; automated test loops, previews and usage-based billing. Public work is sustained — the contribution graph below is live from GitHub.

Public work · sustained

Contact

Building systems that make engineers faster,products smarter,and software more autonomous.

Fasi Owaiz Ahmed

Software Engineer · AI-native developer tools · distributed systems · intelligent infrastructure

Open to software engineering roles·Hyderabad, India · IST

EmailGitHubLinkedInResume

Portfolio v1.0.0 · updated July 2026