Portfolio / 2026 Gurugram, India · IST

Khalid Najam · Applied AI Engineer / SDE III

Khalid builds software for the messy real world.

Khalid builds AI investigation tools and a payroll assistant at PeopleStrong. His work connects models to code, data, and enterprise workflows, with explicit identity, durable state, and controlled actions.

View selected work↓ View résumé↗
5y 2m engineering journey since Jun 2021 Durable state investigations people can continue Scoped identity employee context per request Enterprise depth money, workflow, and compatibility
00 / Point of view

The interesting work starts when the demo ends.

In ATR, an engineer can continue from the same investigation state when automation stops. In Payslip Agent, employee identity belongs to each request. His payroll work brings the same attention to dates, precision, approvals, and compatibility.

CLEAR STATE · USEFUL EVIDENCE · SAFE FAILURE · HUMAN HANDOFF

01 / Selected systems

A few things he has built.

Four case studies: the problem, Khalid’s contribution, a difficult engineering decision, and the limits of the public evidence. Internal systems are described without customer data or proprietary implementation details.

01 Agentic investigation platform · 2026 Internal engineering system

ATR — Automatic Ticket Resolution System

Turn support tickets into evidence-backed diagnoses.

An enterprise investigation system that coordinates specialist agents across support context, source code, data, telemetry, and product knowledge. Teams can bring different capabilities without needing a separate implementation for each one.

Teams capability-aware workspaces HITL continuation from the same trace Graph context across repositories
Read ATR case study
02 Secure AI assistant · 2025–2026 Internal payroll assistant

Payslip Agent

Payroll intelligence with identity handled explicitly.

A production payroll assistant for authenticated payslip retrieval, explanations, tax analysis, and protected documents. Its capabilities were later exposed through MCP tools with request-scoped employee context rather than shared mutable user state.

Identity scoped per request SSE streaming responses PII aware logs and protected documents
Read Payslip case study
03 Enterprise product engineering · 2022–2026 Platform contributions

Enterprise Payroll Platform

Years of edge cases, compressed into reliable product behavior.

Full-stack work across loans, investments, NPS, reimbursements, CTC, variable pay, APIs, workflows, calculation engines, migrations, and production incidents. Years of edge cases made reliability a practical discipline rather than a slogan.

Money BigDecimal migration 8 wk NPS V3 delivery JDK 17 platform modernization

Feature and migration contributions within a wider platform.

Read Payroll case study
04 AI research · IIT Kharagpur · 2021–2022 IIT Kharagpur research

Geospatial AI Research

Learning patterns from satellite radar data.

Deep-learning classification of volcanic deformation using Sentinel-1 SAR and PSInSAR data, comparing AlexNet, ResNet, and autoencoder approaches with augmentation and transfer learning.

CNN architecture comparison SAR Sentinel-1 imagery CNN architecture benchmarking
Read Research case study
02 / Experience

Built through the stack.

One company, progressively harder problems: full-stack payroll, platform modernization, secure AI products, and now agent harness engineering.

Engineering journey5 years 2 monthssince June 2021
May 2026 — Present
PeopleStrong

Software Development Engineer III

Own applied-AI and harness engineering for ATR: agent design, evidence tools, team onboarding, operator workflows, guardrails, production recovery, outcome reporting, and the controlled path from diagnosis to action.

Agent harnessesTypeScriptOpenCode SDKPostgreSQLMCPObservability
Nov 2024 — Apr 2026
PeopleStrong

Software Development Engineer II

Built and productized the Payslip Agent with FastAPI, LangChain, Gemini, Redis, Elasticsearch, secure document retrieval, streaming responses, tax tooling, PII-aware logs, and later MCP integration.

GeminiLangChainFastAPIRedisElasticsearchJDK 17
Jul 2022 — Nov 2024
PeopleStrong

Software Development Engineer

Built major enterprise payroll releases across Loan, Investment V3, NPS V3, BRE, CTC, Variable Pay, and FBP — across entity, service, REST, workflow, SQL, and Angular layers.

Java EEAngularHibernateSQL ServerRESTWorkflow
Jan 2022 — Jul 2022
PeopleStrong

Software Engineer Intern

Built loan product configuration, eligibility and EMI rules, approval workflows, and employee / approver portals. Converted the internship into a pre-placement offer.

JavaAngularPayroll
Jun 2021 — Jan 2022
IIT Kharagpur

Research Intern — Artificial Intelligence

Developed volcanic-deformation classifiers using Sentinel-1 SAR imagery and PSInSAR data under Prof. Debashish Chakravarty, comparing multiple CNN architectures and reporting 90%+ accuracy.

PyTorchCNNsRemote sensingResearch
03 / Working set

The stack is the system around the model.

A useful stack is not a logo wall. Each layer has a job: make intelligence usable, keep state honest, expose evidence, and make the production system operable.

01AI executionreasoning layer

The model is useful only when the loop around it is reliable.

ATR uses specialist investigation stages, tool access, and explicit review gates. Payslip Agent connects retrieval and calculation tools to authenticated employee requests.

WHAT LIVES HEREtool contracts · evidence retrieval · agent state · evaluation · guarded actions
OpenAIGeminiAgent runtimesMCPRAGKnowledge graphsEvaluation
02Application boundaryproduct layer

Intelligence becomes a product at the interface boundary.

FastAPI and streaming responses support Payslip Agent. Java services, REST APIs, and Angular interfaces carry payroll features from data model to employee and administrator workflows.

DESIGN QUESTIONWhere should a failure become visible, and who needs enough context to recover?
TypeScriptPythonJavaFastAPIExpressAngularReactRESTSSE
03State & knowledgememory layer

State is easier to reason about when its lifetime is explicit.

Request-scoped identity, expiring Redis sessions, durable investigations, and relational payroll history have different owners and lifetimes. Khalid keeps those boundaries explicit.

PostgreSQLSQL ServerRedisElasticsearchGraph contextObject storage
04Runtime & operationsproduction layer

If the system cannot explain its failure, ownership gets expensive.

Graph-persistence failures, runtime upgrades, and production support inform the delivery work. Container tooling and observability sit alongside the application, not outside its design.

OPERATING PRINCIPLEThe useful question is not “did it fail?” but “can the next engineer see why and continue safely?”
DockerKubernetesGitHub ActionsGrafanaNew RelicPrometheusLoki
A FEW ENGINEERING INSTINCTS

Understandable before clever.

01

Show the evidence

Important decisions should be traceable. If the system cannot support a claim, it should make that uncertainty visible.

02

Design for the environment that exists

Teams rarely have identical tools or access. The software should adapt to those differences instead of pretending they are not there.

03

Make state explicit

Request context, durable run state, and shared runtime state have different lifetimes and should be treated that way.

04

Make handoff boring

When a person takes over, the useful context should already be there. Restarting from a summary is avoidable waste.

04 / About

Enterprise engineer, now building applied AI.

Khalid is an Applied AI Engineer and SDE III at PeopleStrong. His route into AI came through enterprise software, so his questions tend to be practical: Where does identity come from? What survives a retry? Who can write? What happens when a model is wrong? Can an engineer pick up the exact state?

Before agent systems, he spent years building payroll products across Java, Angular, SQL, workflows, calculation engines, integrations, migrations, and production support. Earlier, he worked on deep-learning classification of Sentinel-1 satellite data at IIT Kharagpur.

That combination now shapes his work: technically ambitious systems that are still clear enough for other engineers and operators to understand and trust.

BasedGurugram, India
EducationB.Tech EEE · Minor in Data Analytics · KIIT
Current focusAgent harnesses · enterprise AI · operator workflows
IF A PDF IS EASIER

Prefer the résumé?

There is a one-page version for a quick scan and a two-page version with more technical detail.

Applied AI résumé1 page ↓ Applied AI résumé2 pages ↓
05 / Contact

Applied AI · Engineering · Systems

Hard system problems are worth a conversation.

khalidnajam7@gmail.com↗
GitHub ↗ LinkedIn ↗
Gurugram, India · IST