Het Shah

1. Oct 2026 to now

   Software Engineer 2

   Infocusp Innovations · Ahmedabad

   Part of a team on a client engagement studying an AI agent evaluation pipeline: how agent behaviour can be measured and validated reliably. I bring production experience with multi-agent LLM pipelines and LLM tracing to ground the approach.

2. Dec 2021 to Oct 2026

   Senior Software Engineer

   Crest Data · Ahmedabad

   Architected and delivered AI and application-security platforms for enterprise customers including Armis, Comcast and Splunk, end to end: design, implementation, testing and deployment. Mentored junior engineers on system design, code review and pre-commit tooling. The four projects below come from this role.

3. May to Jul 2021

   Website Developer Intern

   LearnElectronics · Remote

   Built interactive quizzes and event registration portals with FastAPI, HTML, CSS and JavaScript.

Four projects from my time at Crest Data, 2021 to 2026. Each was built for one of its enterprise customers.

• 

  01Multi-tenant AppSec platform

  Event-driven backend for an application-security product. Twelve scanners behind one execution engine, and an ingest path moved from SQS to Kafka to absorb a weekly burst from 250K to 1.5M findings a day.

  32 enterprise tenants

  No burst throttling

  Client

  Armis, through Crest Data

  My role

  Backend architect. I led the ingest migration.

  • Architected the backend as event-driven microservices: FastAPI, with Lambda → SQS → ECS Fargate, autoscaling and dead-letter-queue alerting.
  • Built the engine that runs 12 SAST and DAST scanners, with unified finding normalisation, dependency chain resolution and SBOM/VEX generation.
  • Built an on-demand scan API (S3-backed uploads, SQS-driven ECS) and container image scanning across AWS ECR, GCP GCR and Docker Hub.
  • Built JWT and Basic auth middleware with rate limiting and cross-tenant isolation.
  • Designed a Valkey cache holding about 4M live keys at an 81% average hit rate, 95%+ during full scans.
  • Led the move of findings ingest from SQS to Kafka. It decoupled ingest throughput from the datastore's write capacity and ended burst throttling.

  Python, FastAPI, AWS (ECS Fargate, Lambda, SQS, S3, DocumentDB, Valkey, ALB, ECR, EventBridge, Secrets Manager, CloudWatch), Kafka, Docker

  Application Security

• 

  02Five-stage LLM fix pipeline

  A multi-agent pipeline that finds vulnerabilities and proposes validated fixes. Two model providers behind one abstraction, every call traced, and 85% of prompt tokens served from cache.

  212K calls a day

  98.5% success

  Client

  Armis, through Crest Data

  My role

  Built the provider abstraction and the tracing.

  • Five stages: AST-based code scanning with tree-sitter, cross-scanner deduplication, reachability validation with Joern, agentic fix proposal, and fix validation.
  • Built one abstraction over Cerebras and AWS Bedrock, with a model configured per stage and structured output extraction with retries.
  • Integrated Langfuse so every LLM call across the pipeline is traced.
  • It powered automated vulnerability detection and fix generation for the platform in project 01.

  Python, Cerebras, AWS Bedrock, Langfuse, tree-sitter, Joern, Pytest

  AI Systems

• 

  03Security event ingestion

  Ingestion frameworks and pipelines that bring new sources into a security data platform, then normalise and route every event with retries built in.

  Security data platform

  2,000 events a minute

  Client

  Comcast, through Crest Data

  My role

  Built the ingestion frameworks and pipelines.

  • Designed API ingestion frameworks and integrated new security data sources into the platform.
  • Built pipelines in Benthos and Bloblang that normalise and route security events at about 2,000 events a minute.
  • Integrated external scanning services with error handling and retries.

  Python, Benthos, Bloblang, REST APIs

  Data Pipelines

• 

  04Observability tooling

  REST APIs and dashboards for an enterprise observability suite: reusable content modules, a licence upgrade path, and proactive SLA monitoring and alerting.

  Django and React

  Proactive SLA alerts

  Client

  Splunk, through Crest Data

  My role

  Built the APIs, dashboards and alerting.

  • Refactored an application for SAP-specific data visualisation and its Splunk integration.
  • Built a licence upgrade mechanism for a modular Splunk app, and reusable content pack modules for customer deployments.
  • Developed proactive SLA monitoring and alerting pipelines in Splunk's observability suite.

  Python, Django, React, MongoDB, Splunk

  Observability

Enterprise tenants on one platform

32

Findings a day at the weekly peak

1.5M

LLM calls a day at 98.5% success

212K

Prompt tokens served from cache

85%

• Backend and APIs

  Python, FastAPI, Django, Pydantic, REST, JWT auth, rate limiting

  01 AppSec platform 04 Observability tooling

• Distributed and event-driven

  Kafka, SQS/SNS, EventBridge, multi-tenant SaaS, dead-letter queues, caching with Valkey (Redis)

  01 AppSec platform 03 Event ingestion

• AWS

  ECS Fargate, Lambda, SQS, S3, ECR, ALB, EventBridge, Secrets Manager, CloudWatch, VPC, IAM. Infrastructure in Terraform.

  01 AppSec platform

• AI and LLM systems

  Multi-agent pipelines, AWS Bedrock, Cerebras, structured output, prompt caching, RAG, Langfuse tracing, MCP

  02 LLM fix pipeline

• Data pipelines

  Benthos, Bloblang, API ingestion, normalisation and routing

  03 Event ingestion

• Application security tooling

  SAST/DAST, Trivy, Gitleaks, Checkov, OpenGrep, Semgrep, Grype, SBOM/VEX, tree-sitter, Joern

  01 AppSec platform 02 LLM fix pipeline

• Databases

  DocumentDB (MongoDB), DynamoDB, Valkey (Redis)

  01 AppSec platform 04 Observability tooling

• Observability

  CloudWatch, Prometheus, Grafana, Splunk

  01 AppSec platform 04 Observability tooling

• Delivery and testing

  Docker, GitHub Actions, GitLab CI/CD, pre-commit, Pytest, integration and end-to-end suites in containers

  All four projects

Also: JavaScript, React, Bash, C and C++.