Software Engineer

DinaGanai

Building agentic AI systems and production infrastructure. Currently at Okta — RAG pipelines, evaluation frameworks, and autonomous AI tooling.

View work GitHub ↗

embed.py search.py

1# semantic job search engine
2
3async def search(query: str) -> list[Job]:
4  vector = await embed(query)
5  hits = await pinecone.query(
6    vector, top_k=20, filter="remote": True
7  )
8
9  return [Job.from_hit(h) for h in hits]

// experience

Okta

2024 — present

Software Engineer · AI Productivity

Agentic AI RAG Golang Python TypeScript AWS Claude Code

Berkeley Payment Solutions

2022 — 2024

Software Engineer · Fintech Infrastructure

Golang Elixir Phoenix React Kubernetes AWS PCI Compliance

// selected projects

Internal AI Assistant Okta work A first-responder bot for internal support tickets — webhook on create, agent pulls docs and past tickets, posts analysis before a human triages.

Internal AI Assistantlive · okta

▾ internal-ai-assistant

  webhook.py

  agent_client.py

  responder.py

# webhook.py · Jira webhook handler
 
from flask import Flask, request
import agent_client, jira
 
app = Flask(__name__)
 
@app.route("/webhook", methods=["POST"])
def handle_ticket():
  payload   = request.json
  ticket_id = payload["issue"]["key"]
  summary   = payload["issue"]["fields"]["summary"]
  body      = payload["issue"]["fields"]["description"] or ""
 
  response = agent_client.query(ticket_id, summary, body)
  if response:
    jira.post_comment(ticket_id, response)
  return {"status": "ok"}, 200

A first-responder bot for internal support tickets. A Jira webhook fires on ticket creation, sends the details to an enterprise AI agent that queries connected internal documentation and past tickets, and posts its analysis back as a comment — surfacing relevant context before a human even looks at it. Built the Python integration layer: webhook handler, API client, response parser, and Slack alerts for failures. Meaningfully reduced triage time for the support team.

KYC Compliance Automation Berkeley Payment Solutions work Automated a manual KYC process that took hours each month — Elixir publishes to SQS, Go Lambda screens against sanctions lists, stores encrypted results.

KYC Compliance Automationlive · berkeley payment

▾ kyc-compliance-automation

  handler.go

  screener.go

  screening_queue.ex

// handler.go · Lambda entry point
 
func Handler(ctx context.Context, event events.SQSEvent) error {
  db, err := connectDB()
  if err != nil {
    return fmt.Errorf("db connect: %w", err)
  }
  defer db.Close()
 
  for _, msg := range event.Records {
    var req ScreeningRequest
    if err := json.Unmarshal([]byte(msg.Body), &req); err != nil {
      return fmt.Errorf("parse message: %w", err)
    }
    result, err := screen(ctx, req)
    if err != nil {
      return err
    }
    if err := storeEncrypted(ctx, db, result); err != nil {
      return err
    }
  }
  return nil
}

Automated a manual KYC compliance process that previously required hours of work per month. Elixir frontend publishes screening requests into SQS; Go Lambda consumes them, checks customer data against sanctions lists, and stores encrypted results in Postgres. Replaced an error-prone manual workflow with a reliable, auditable pipeline — meaningfully reducing compliance team overhead.

mylevain personal product A baking companion — starter health, baker's math, bake-day replanning, and vision on crumb structure — multi-agent tool loop I use every week.

A baking companion that tracks starter health, adapts recipes with baker's math, orchestrates bake day with dynamic replanning, and uses vision to analyze crumb structure — three cooperating agents and a custom tool loop I use every week.

Agentic AINext.jsAnthropic SDKComputer visionTool useFull stackSourdough

landing ↗ preview ↗ read the write-up ↗

Repository is private.

// writing

Context efficiency in Claude Code: the art of saying less to do more

Part 2: why context bloat kills productivity, a three-tier documentation model, deduplication rules, and a playbook to reclaim tokens—without drowning your team in docs.

Mar 2026

featured

Making repositories more AI-proficient with Claude Code

A four-layer framework for context engineering: CLAUDE.md as a memory anchor, skills, scaffolding, and MCP—plus workflows that keep sessions honest.

Mar 2026

View all →

// about

the work

I'm a Software Engineer on the AI Engineering Productivity team at Okta, where I build agentic systems, RAG pipelines, and AI tooling that makes engineers faster. AI isn't just the job — it's what I'm genuinely excited about.

I got here the non-linear way — a career change (more like lack of), a bootcamp that has since gone out of business (RIP), and a lot of figuring it out as I went. Shoutout to my dad and my brother for that one. Turns out it's a pretty common story in tech.

agents · RAG · eval · Python · TypeScript · Go

outside of it

My biggest passion is baking. I'm the designated family baker — all the desserts, all the holidays, no complaints. Recently I've gotten deep into sourdough and made my own starter from scratch with, yes, AI assistance. She's thriving.

I also spend a lot of time at the gym (lifting), in hot yoga and pilates, and getting my steps in. My dad recently signed me up for a 5k in April without asking. Wish me luck. And my nieces — they're the best, full stop.

sourdough · lifting · hot yoga · family

// contact

Open to interesting conversations — reach out however works best.

email dinaganai@gmail.com github github.com/dganai linkedin linkedin.com/in/dinaganai