01Working build
Agentic Data Reliability
Self-Healing Data Pipeline Agent
An agentic ETL system that detects schema drift, drafts controlled fixes, requires human approval, validates recovery, and automatically rolls back failed mutations.
AI student · builder · currently exploring internship opportunities
AI ENGINEERING · AGENTIC SYSTEMS · CLOUD
I build AI systems that reason, act, and stay accountable.
I’m Chahak Goswami, a Penn State Artificial Intelligence student focused on practical agentic AI, trustworthy automation, cloud engineering, and turning ambiguous problems into working prototypes.
4agentic AI projects
FAAfederal AI internship
40%token cost reduction contribution
3.70Penn State GPA
SELECTED WORK
Systems over slideware.
I like projects where AI has to do more than generate text: reason across evidence, interact with tools, make bounded decisions, recover from failure, and leave a trace.
02Working build
Trustworthy AI Research
Deep Research Agent with Citation Graph
A multi-hop research agent that expands questions, retrieves evidence, builds a citation graph, grades sources, detects contradictions, and produces a traceable report.
03Project build / roadmap
Agentic Operations
Autonomous Ticket Resolution Engine
An agentic support workflow designed to read tickets, inspect data, plan safe versus destructive actions, request confirmation where needed, and produce auditable resolution records.
04Project build / roadmap
AI for Developer Workflows
CI Triage Agent
An engineering agent designed to read failing CI logs, classify errors, reproduce failures, propose code patches, validate fixes, and open approval-ready pull requests.
EXPERIENCE
Building AI in a real federal environment.
At the FAA / Rigil Corporation, I worked across RAG, agentic workflows, speech AI, cloud security visibility, and AI FinOps.
See experience2026 · Washington, D.C.
Agentic AI Intern
Federal Aviation Administration / Rigil Corporation
- Built FAAGPT with Bedrock Knowledge Bases and Claude.
- Designed FAA Airspace Agents with AgentCore, Claude Agent SDK, LangChain, and LangGraph.
- Built traceable multi-agent workflows and supported model quality evaluation.
- Reviewed AI/cloud usage and contributed to a 40% token-cost reduction.
TOOLKIT
What I build with.
My focus is not collecting logos. These are technologies I use to build, test, debug, and reason about AI systems.
Agentic AI
AI / GenAI
Engineering
Cloud / Security
HOW I THINK
“The interesting part of autonomous AI isn’t just what it can do. It’s how clearly we can define what it should do, what it must never do, and how we prove the difference.”
That idea shows up across my projects: approval gates, rollback paths, source provenance, contradiction checks, audit trails, and deterministic tests.