Full Stack Software Engineer · Lahore, PK · Remote

I build the infrastructure
behind AI products.

Five years shipping production systems, now focused on LLM infrastructure. At PromptLayer I build tool-calling runtimes, MCP servers and agent observability used by 100+ AI teams. Before that, four years of Rails, React and Postgres, which is why wiring AI into existing systems, not just greenfield demos, is where I'm most useful.

See the work Get in touch Résumé
run · portfolio.visit
    idle click a span to jump
    0yrsshipping production software
    0+LLM providers in one agent runtime
    0+API tools exposed via a public MCP server
    0+AI teams using features I shipped end to end
    span · get_profile()

    About

    I'm Usama, a software engineer from Lahore who likes the unglamorous middle of AI products: the runtime that turns a model's tool call into a real request, the trace that explains why a run failed, the schema that keeps 60 tools honest.

    My path ran through four years of consulting at Devsinc, building Rails and Python systems for US startups, then into LLM infrastructure at PromptLayer. That order matters. I've debugged enough Postgres migrations and Celery queues to know that the hard part of "add AI" is rarely the model.

    Off the clock I write, read, game, and hunt for new food. I care about AI that lands well for people, and about data privacy as a feature rather than a footnote.

    • now Full Stack Engineer @ PromptLayer
    • edu BS Computer Science, FAST NUCES '21
    • tz PKT (UTC+5), overlapping US mornings
    • oss mcp-audit · why-did-you-fetch
    Portrait of Usama Anser
    usama.anser · v5.0
    span · list_experience()

    Experience

    Rendered as a span waterfall, because that's how I read everything else.

    01/2026 — present USA · remote

    Full Stack Software Engineer @ PromptLayer

    Prompt engineering and LLM observability platform that AI teams use to manage, version, evaluate and monitor prompts in production.

    • Shipped 10+ major features across a Python/FastAPI + React/TypeScript stack, owning each from Postgres schema to UI, used by 100+ teams.
    • Built the Tool Registry from scratch: versioned, reusable tool definitions with an editor UI, public API, RBAC and response-schema enforcement.
    • Engineered the agentic auto-execution loop, a multi-provider engine driving LLM-to-tool round trips across streaming and non-streaming paths, with per-call OpenTelemetry spans across 10+ providers.
    • Added AI-mocked agent runs so users can test full loops without live endpoints, cutting agent iteration time by 50%.
    • Built PromptLayer's official MCP server, exposing 60+ API tools to Claude, Cursor and other MCP clients.
    • First-party support for provider-native tools (web search, code execution, file search, bash) across Anthropic, OpenAI, Google, Vertex AI and Bedrock.
    • AI-generated trace summaries for agent runs, cutting time-to-diagnose a failed run by 70%.
    • Automated weekly release notes with PromptLayer's own workflow product, replacing a 3-hour manual process.
    PythonFastAPIReactTypeScriptPostgresOpenTelemetryMCP
    02/2024 — 01/2026 USA · remote

    Senior Software Engineer @ Devsinc

    IT services and consulting, building scalable web applications for startups and enterprises across the US, Europe and the Middle East.

    • Built an LLM-powered lead engagement pipeline end to end: GA4-instrumented forms → SFTP → Celery workers → AWS Glue ETL → model layer → BigQuery.
    • Designed the generation stage: enriched behavioural data with location and profile signals, fed structured context to an LLM producing personalised outreach.
    • Orchestrated the async pipeline with Google Cloud DAGs across ingestion, transformation, inference and warehouse load.
    • Integrated the Microsoft ecosystem (Office 365, Azure) via Microsoft Graph API, boosting customer engagement by 30%.
    • Introduced Trailblazer for modular, reusable Rails code.
    PythonCeleryAWS GlueGCPBigQueryRailsMS Graph
    11/2021 — 02/2024 USA · remote

    Software Engineer @ Devsinc

    • Designed a Python web app with auth, authorization and real-time WebSockets, boosting engagement by 35%.
    • Built a React app with RxJS for real-time streaming and state management.
    • Distributed Redis caching for a Rails app, improving response times by 45%.
    • Integrated Elasticsearch for advanced search and faster retrieval.
    PythonDjangoRailsReactRxJSRedisElasticsearch
    09/2017 — 05/2021 Lahore, PK

    BS Computer Science @ FAST NUCES

    National University of Computer & Emerging Sciences.

    span · list_projects()

    Selected work

    Open source tools, platform work and a few full-stack builds.

    open source github ↗

    mcp-audit

    Static security analysis for MCP servers written in Python. Point it at a repo and get findings for tools exposed without auth, static-key comparisons, SSRF-prone URL handling and tool input reaching shells, eval or raw SQL. Standard-library AST only, with SARIF output for GitHub Code Scanning.

    $ mcp-audit server.py
server.py:11:12  CRITICAL  MCP004  Tool input reaches a dangerous sink
server.py:9      HIGH      MCP001  Tools exposed without authentication
2 findings · 1 file · 1 tool
    PythonASTSARIFSecurity
    open source github ↗

    why-did-you-fetch

    Dev-only devtool that finds duplicate requests, request waterfalls and runaway refetches, and names the React component that caused them. Zero config, zero deps, compiles out of prod.

    TypeScriptReactDevTools
    promptlayer

    Tool Registry & agent loop

    Versioned tool definitions with an editor, public API and RBAC, plus the multi-provider auto-execution engine that runs LLM↔tool round trips with OTEL spans on every call.

    FastAPIReactOpenTelemetry10+ providers
    promptlayer

    Official MCP server

    60+ PromptLayer API tools exposed to Claude, Cursor and any MCP client, with JSON response schemas and structured outputs that render charts as images.

    MCPPythonSlack OAuth
    product

    Book Pedia

    AI reading companion with in-depth chapter summaries and Q&A over your books. Retrieval-augmented generation on pgvector embeddings.

    FastAPIpgvectorRAGOpenAI
    product

    Lead Engagement Platform

    LLM-driven personalisation pipeline: GA4 forms → async Celery workers → AWS Glue ETL → personalised outreach → BigQuery.

    PythonCeleryGCP DAGsBigQuery
    product

    Meal Suite

    Diet and medication tracking for seniors with daily, weekly and monthly reports shared with doctors and guardians.

    DjangoReactCeleryAWS
    product

    CRM365

    Customer insights CRM spanning sales, service and marketing automation, integrated deep into Microsoft 365 via Graph API.

    PythonReactPostgresMS Graph

    More on GitHub ↗, including Rails, React and data-visualisation side projects.

    span · get_skills()

    Toolbox

    ai / llm

    Anthropic APIOpenAI APIGeminiBedrockMCPAI AgentsTool callingRAGEmbeddingspgvectorLangChainLangSmithLLM evalsLLM observabilityPrompt managementFine-tuning

    backend

    PythonFastAPIDjangoRubyRailsCeleryRESTWebSocketsPytest

    frontend

    TypeScriptJavaScriptReactNext.jsRxJSTailwind

    data & infra

    PostgreSQLRedisElasticsearchBigQueryAWSEC2 / S3AWS GlueGCPAirflowETL pipelinesDockerKubernetesCI/CDOpenTelemetry
    span · respond()

    Let's build something
    that survives production.

    Open to full-time and contract roles on AI infrastructure, agent tooling and full-stack product work. Remote-first, comfortable overlapping US hours.

    email work.usama.anser@gmail.com github UsamaAnser1518 ↗ linkedin usama-anser ↗ résumé Download PDF ↓