Skip to content
View salehaiftikharr's full-sized avatar

Block or report salehaiftikharr

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
salehaiftikharr/README.md

Saleha Iftikhar

I build production web apps and autonomous AI agents, end to end.
CS @ Gettysburg College · I like shipping things people actually use, and making AI reliable enough to trust.

Portfolio LinkedIn Email


🚀 What I'm building

Saaya  ·  a persistent AI coworker, not a chatbot. You hand it real jobs and it keeps them alive across crashes and restarts, risky commands stop at approval gates until a human signs off, and its long-term memory keeps sources and versions for everything it learns. One coworker across web, Slack, and MCP, built on LangChain Deep Agents and LangGraph. Live demo (passphrase gated to keep model costs sane, ask me for the key).

Forge & the Minions  ·  an agent that builds and self-tests other agents, now a live web app. Describe the specialist you need in plain English and Forge compiles it into a runnable, tool-using agent, then repairs it through a build, run, judge, and refine loop. Its autonomous "minions" pick up a GitHub or Linear issue, fix it, run the tests in a sandbox, and open a pull request only when a verification gate passes: zero unsafe ships on a hand-labeled eval, with run tracing, trajectory-level evals, and a memory of past repairs.

Grounded  ·  a retrieval Q&A system that guards both ends. It refuses to answer when retrieval is too weak, and a separate model checks every claim in the answer against its sources and flags anything unsupported. Live demo.

Analytics Chat Assistant  ·  ask a question in plain English, get back validation-gated, read-only SQL and streamed charts. Provider swaps between Claude and GPT, graded by a 25-case eval harness.

Application Tracker  ·  a React and Supabase job tracker with a Chrome (MV3) extension that reads six ATS platforms into a profile-based autofill engine.

Fuerza  ·  a full-stack social fitness app (feed, profiles, kudos, comments, messaging), deployed live.

🛠️ Tech I reach for

Also: autonomous agents · LLM evals and eval harnesses · MCP · LangChain Deep Agents · agentic coding with Claude Code

🌱 A little more

I got into engineering by building all sorts of things, from a small home baking business to data dashboards to policy research with the Eisenhower Institute. What stuck was the habit of shipping real things and caring about the parts that tend to break. Lately I am most drawn to agents that are genuinely trustworthy, not just impressive in a demo.

Reach out if you are building something in this space, or think I would be a good fit on your team. I am always glad to connect.

Pinned Loading

  1. saaya saaya Public

    A persistent AI coworker on LangChain Deep Agents and LangGraph: durable jobs with approval gates, memory with provenance, restart-proof state, one coworker across web, Slack, and MCP.

    TypeScript

  2. agent-forge agent-forge Public

    Plain English becomes inspectable specialist agents (Minions) that fix real issues and open a PR only when a test proves the fix, or decline. CLI, Slack, and a web product on one verification-gated…

    TypeScript

  3. analytics-chat-assistant analytics-chat-assistant Public

    Chat-native BI tool: natural-language questions become validation-gated single-SELECT SQL and streamed charts, over a read-only Postgres role, with a swap-in Claude/GPT provider and a 25-case eval …

    TypeScript

  4. grounded grounded Public

    A retrieval Q&A agent that cites its sources and refuses when the answer is not grounded in the corpus. TypeScript, Vercel AI SDK, with a grounding gate and an eval harness.

    TypeScript

  5. fuerza fuerza Public

    Strava for lifters — a full-stack social fitness app (workout feed, profiles, kudos, comments, messaging). React + Node + MySQL, deployed live with a one-click, no-backend demo mode.

    JavaScript

  6. DataVisualizationPortfolio DataVisualizationPortfolio Public

    Portfolio of projects exploring topics I care about - including women in tech, programming trends, and wage disparities - created using tools like Python (Pandas, Matplotlib) and Plotly.

    Jupyter Notebook