Forward Deployed Engineer at Exotel · CSE '26, IIIT Dharwad

, I'm Aryan, a software engineer.

I build software that ships to production.

Forward Deployed Engineer at Exotel and a CS grad from IIIT Dharwad ('26). I've shipped LLM voice agents to production, built the eval platform behind them, wired the integrations, automated the pipelines, and built full-stack products on the side.

Open to new roles · anywhere in India or remote

Download resume Copy email

GitHub LinkedIn

Highlights

The short version.

aryan@macbook: ~

❯ whoami

aryan jaiswal · software engineer · cse '26, iiit dharwad · bangalore

❯ git log --author=aryan --oneline -- argus/ | wc -l

500+ commits · 79 merged PRs, and counting

❯ ./autoloop --stages

evaluate → investigate → recommend → [approve] → apply → verify → compare → journey

❯ ls ~/shipped

argus/ autoloop/ voice-sim/ booking-routes/ campaign-pipeline/ the-wedding-home/ aura/

❯ cat status

● open to new roles · anywhere in india or remote

❯

Experience & education

Where I'm shipping now.

Exotel, a voice AI platform. I joined as a GenAI intern and converted to full-time.

Exotel

Voice AI / conversational AI platform · Bangalore

Forward Deployed Engineer
Jun 2026 – Present
- Shipped 10+ LLM voice agents to production in 5 languages and took 16 bots for 15+ clients through formal evaluation, owning requirements, prompts, tool/KB design, integrations and go-live.
- Ran client calls, demos and requirement sessions with CSM and sales across 10+ accounts, turning call feedback into same-day fixes.
- Built AutoLoop inside Argus: a 7-stage workflow in an agentic harness with symlinked bot repos and 16 skills. It generates scenarios, compares prompt techniques in chat simulations, selects winners and rolls back regressions. Case study
- Built Argus's in-house voice simulator for tests between chat simulations and real calls; made Argus/AutoLoop usable from AI coding tools (Cursor, Claude Code). Harness used on 10+ builds; bot delivery went from ~1 week to 1–2 days (estimate). Case study
- Migrated Bawa's booking agent for 6 hotels to the CIN booking provider with HMAC-signed Apache Camel routes; passed 313/313 route and 114/114 booking checks on staging, then confirmed bookings after the production cutover. Case study
- Automated Kinetic Green's campaign cycles (150K lead entries) with Python, Jenkins and AWS S3/SES: post-call lead reports and deterministic PDF summaries emailed after each campaign, with backups and validated lead reloads. Case study
Converted to full-time ↑
Tech Intern, GenAI
Oct 2025 – Jun 2026
- Built Argus, a voicebot debugging/evaluation platform with Google Chat access, Gemini tool calling, VAD latency analysis, an HMAC-signed recording proxy and Splunk On-Call alerts; cut triage from ~30 min to under a minute. Case study
- Hardened the LangGraph LLM path as DRI for 150K+ calls/month (incl. BlueDart): 7 s wall-clock timeout/fallback, parallel-tool-call guards, TTS markdown filter and a reasoning-leak guard tested on 269 production replies. Case study
- Scaled Pipecat WebSocket handling from 1 to 8 workers behind NGINX least_conn; added Prometheus/Grafana monitoring with EC2 discovery and 5 Ansible roles for provisioning. Case study
- Added per-company custom test cases and scenarios to the bot test framework: FastAPI endpoints, Redis storage, a React/TypeScript UI and a scenario-mix algorithm.

EarlierJoveo · InfoKalash · Infosys Springboard

Joveo · AI/ML Intern, Supply Automation
Oct 2025 · 2 weeks
Built an email-processing and enterprise-search pipeline with n8n, Gemini embeddings and PostgreSQL/pgvector, with entity extraction, LLM summaries and WhatsApp/Telegram alerts.
InfoKalash · UI/UX Developer Intern
Jan 2025
Redesigned navigation and Figma workflows through research with 15+ users, producing reusable components, interactive prototypes and developer handoffs.
Infosys Springboard · AI/ML Intern
Mar – Jun 2024
Built a CNN plant-disease classifier with Python and Flask, including image preprocessing and model-serving workflows. Presented at the inaugural Infosys Springboard Summit.

IIIT Dharwad · B.Tech, Computer Science and Engineering (2022–2026)

CGPA 7.76/10 · Graduated May 2026

Gold Microsoft Student Ambassador: founded the MSA chapter (500+ members)
Inaugural Alumni Secretary: set up the college's alumni community

Data Structures and AlgorithmsOperating SystemsDBMSComputer NetworksOOP

How I ship

From customer call to production.

Six steps, each shown with a real artifact from a different piece of work.

01
Start on the customer call. Write down what the agent must do, in their words, before touching code.
From Hotel-booking migration
requirements.md
- Six properties, English + Hindi callers
- Move booking to the new provider without breaking live calls
- Payment link by invoice; guard against duplicate bookings in a session
- Cut over only after the staging harness is green
02
Draw the routes and the failure paths first: signing, retries, timeouts and what happens when a call repeats.
From Hotel-booking migration
4 Apache Camel routes
1. 1Session setup
2. 2Availability
3. 3Booking + guard
4. 4Payment-link invoice
03
Small, typed, boring code at the edges. Every outbound request is signed; every retry is bounded.
From Hotel-booking migration
sign.ts
```
const signature = createHmac("sha256", secret)
  .update(`${timestamp}.${body}`)
  .digest("hex");

await withRetry(() => post(url, body, {
  "X-Signature": signature,
  "X-Timestamp": timestamp,
}), { attempts: 3 });
```
Illustrative sketch of the HMAC-SHA256 signing step, not the production code.
04
Prove it on staging before anyone hears it on a call. Chat simulations, then voice simulations, then real calls.
From Argus / AutoLoop
staging harness
- route comparisons313/313 passed
- booking checks114/114 passed
- 14-scenario regression, local run time (vs ~20 min remote)~1 min
05
Automate the boring, dangerous parts. Integrity checks block destructive steps; dry-run mode comes first.
From Campaign automation
Jenkins · twice daily
1. stage 1Dump + reconcile
2. stage 2Reports
3. stage 3Archive + email
4. stage 4Integrity check
5. stage 5Load batch
6. stage 6Compare to source
7/7 successful recorded production builds in the first observation window
06
If you can't see it, you can't scale it. Workers, exporters and dashboards that find new hosts on their own.
From Production infra
WebSocket workers · NGINX least_conn
Worker count 1 → 8 · Prometheus + Grafana · EC2 discovery

requirements.md

Six properties, English + Hindi callers
Move booking to the new provider without breaking live calls
Payment link by invoice; guard against duplicate bookings in a session
Cut over only after the staging harness is green

4 Apache Camel routes

1Session setup
2Availability
3Booking + guard
4Payment-link invoice

sign.ts

const signature = createHmac("sha256", secret)
  .update(`${timestamp}.${body}`)
  .digest("hex");

await withRetry(() => post(url, body, {
  "X-Signature": signature,
  "X-Timestamp": timestamp,
}), { attempts: 3 });

Illustrative sketch of the HMAC-SHA256 signing step, not the production code.

staging harness

route comparisons313/313 passed
booking checks114/114 passed
14-scenario regression, local run time (vs ~20 min remote)~1 min

Jenkins · twice daily

stage 1Dump + reconcile
stage 2Reports
stage 3Archive + email
stage 4Integrity check
stage 5Load batch
stage 6Compare to source

7/7 successful recorded production builds in the first observation window

WebSocket workers · NGINX least_conn

Worker count 1 → 8 · Prometheus + Grafana · EC2 discovery

Selected work

Five things I could talk about for an hour.

AI · Evaluation platform
Making voice bots debuggable, then making them improve themselves
Built from zero to production in three months: triage from ~30 minutes to under a minute, and per-bot delivery from about a week to 1–2 days.
AIEvaluationPython
Chat sim
Voice sim
Real calls
Backend · Integrations
The plumbing behind a voice agent
A signed, guarded hotel-booking migration that passed 313/313 route checks in staging, and a campaign pipeline that turned 25-minute loads into 2.
BackendApache CamelJenkins
DevOps · Infrastructure
Running it in production: workers, monitoring and pipelines
Media workers scaled from 1 to 8 behind NGINX, with monitoring that finds new hosts by itself and pipelines that refuse to do damage.
DevOpsNGINXGrafana
AI · Full-stack
AURA: a five-agent finance advisor
Google Cloud Agentic AI Day finalist, top 3.5% of 57,000+ participants, as team lead.
AIFull-stackMCP
Live site
Full-stack · Web
The Wedding Home: a planning platform for Indian weddings
61 API routes with typecheck, 11 Vitest suites and a Playwright journey in CI on every push.
Full-stackNext.jsCI
Live site

More projects

Built on the side.

AI Career Navigator (GitHub)
Azure OpenAI career assistant: resume feedback, gap analysis, generated mock interviews.
Microsoft JS AI Build-a-thon — Overall Winner + People's Choice
GitHub for AI Career Navigator Demo for AI Career Navigator Microsoft blog for AI Career Navigator
Myriox (live site)
LLM-directed personas with five mobility types walk a floor plan to find accessibility issues, grounded in ADA clauses via vector search.
Top 50 of 2,989 · OpenAI × NamasteDev hackathon
Live for Myriox GitHub for Myriox Demo for Myriox
Cognitive-OS (GitHub)
GitHub-synced developer focus tracker: 0–100 cognitive-load heuristic with z-score anomaly detection, Redis cache, three Azure OpenAI agents, Docker → Azure Container Apps via GitHub Actions.
GitHub for Cognitive-OS
DevTinder (GitHub)
MERN developer-networking app: connection requests, real-time chat, Google OAuth, Razorpay webhooks, Docker.
GitHub for DevTinder Demo for DevTinder
LEELA (live site)
Turns topics, notes or code into narrated short films with timed speech and FFmpeg.
Live for LEELA GitHub for LEELA
Digit Motor FNOL Voice Agent (GitHub)
TypeScript/Fastify insurance-claim voice flow with confirmation gates, bearer auth, an idempotency store, 6 tool schemas and 27 tests.
GitHub for Digit Motor FNOL Voice Agent

All projects

Tech specs

What I build with.

Every item lists where I used it. Hover or focus one to light up that work across the page.

Languages: PythonArgus / AutoLoop · Integrations · Myriox
TypeScriptThe Wedding Home · Digit Motor FNOL · Exotel · This site
JavaScriptAURA · DevTinder · AI Career Navigator
JavaIntegrations
C++Problem solving
SQLJoveo
Backend & integration: FastAPIArgus / AutoLoop · Exotel · Myriox
Node.js / ExpressAURA · DevTinder
FastifyDigit Motor FNOL
Spring Boot + Apache CamelIntegrations
RESTIntegrations · Argus / AutoLoop
WebSocketsProduction infra · Myriox
Socket.IOAURA · DevTinder
HMAC signingIntegrations · Argus / AutoLoop · The Wedding Home
RedisExotel · Argus / AutoLoop · Cognitive-OS
AI / LLM: LangGraphProduction infra · Argus / AutoLoop
LangChainDaily driver
LangflowDaily driver
n8nJoveo
GeminiArgus / AutoLoop · Joveo
Azure OpenAIAURA · AI Career Navigator · Cognitive-OS
MCPAURA · Myriox
PipecatProduction infra
Prompt evaluation & simulationArgus / AutoLoop
RAG (keyword + vector)AURA · Myriox · Joveo
AI coding tools: Claude CodeArgus / AutoLoop · Exotel · This site
CursorArgus / AutoLoop · Exotel
CodexMyriox
Frontend: ReactExotel · DevTinder
Next.jsThe Wedding Home · Myriox · This site
Tailwind CSSDevTinder · This site
FigmaInfoKalash
Cloud & DevOps: AWS (S3, SES, EC2)Integrations · Production infra
Azure (App Service, Container Apps)AURA · Cognitive-OS · Myriox
DockerCognitive-OS · DevTinder
JenkinsIntegrations
GitHub ActionsCognitive-OS
NGINXProduction infra
Prometheus / GrafanaProduction infra
AnsibleProduction infra
VercelThe Wedding Home · This site
LinuxDaily driver
Data: PostgreSQL / pgvectorJoveo
MongoDBThe Wedding Home · DevTinder
OpenSearchArgus / AutoLoop
Testing: PlaywrightThe Wedding Home · This site
VitestThe Wedding Home
PostmanDaily driver

Recognition

Outside the day job.

Paper · COMSNETS 2026
A Bid-Driven Optimization Framework for Real-Time Electric Vehicle Charging Scheduling
Main conference, pp. 60–66. With Pranay Ch., Vinay Jain, Malay Kumar.
IEEE Xplore publication
Microsoft JS AI Build-a-thon
Overall Winner + People's Choice
Troop Lead, mentoring 80+ developers.
Program Badge
Google Cloud Agentic AI Day
Finalist, team lead
Top 3.5% of 57,000+ participants.
Proof
Microsoft
Gold Microsoft Student Ambassador
Highest tier. Founded the IIIT Dharwad chapter, 500+ members. Also a Gold Startup Scout: referred 8 startups.
Microsoft feature
Problem solving
867 LeetCode problems (165 hard)
October 2026 snapshot. GeeksforGeeks institute rank 2 at IIIT Dharwad.
LeetCode GeeksforGeeks
Certified
Microsoft Azure AI Apps & Agents Developer Associate
Also GitHub Agentic AI Developer and Azure AI Fundamentals (AI-900).
Azure AI Apps & Agents GitHub Agentic AI AI-900

Co-organized GitHub Copilot Dev Days Bengaluru with TechNexus at Microsoft Reactor (~200 attendees)
Exotel Swag Battle Winner 2025
Talk: Future Voices: Generative AI Impact on Student Careers
3rd place, Parsec 4.0 BitHunt 2.0, IIT Dharwad

Off the clock