Hajik Renaa

Available2 project slots open this month

I build AI that picks up the phone, reads the paperwork and runs the busywork.

I’m Hajik, a freelance AI & automation engineer. Voice agents, document AI and automations for businesses buried in calls, PDFs and copy-paste. Built end to end, measured, and live in weeks, not quarters.

7+
partner companies
30+
systems shipped
11
languages on live calls
Hajik Renaa, AI & Automation Engineer
Hajik RenaaAI & Automation Engineer · ChennaiOpen to work
Shipped into7+ partner companies
RecruitmentLegal-techMaritime complianceManufacturingElectronics distributionTextile tradeFitness SaaSEducationJewellery retail
OpenAI RealtimeGemini LiveClauden8nWhatsApp Cloud APITwilioVobizElevenLabsSarvam AIDeepgramChromaDBPostgreSQLFastAPINext.jsFlutterDocker

01Listen in

A real call flow, line by line.

This is the kind of call my voice agents make every day. Scroll to play it.

  1. 01

    Speaks the caller's language

    Natural Tanglish, not translated English. Backchannels like ‘seri’ and ‘aama’ don't interrupt it.

  2. 02

    Answers from your knowledge base

    Salary bands, locations, policies: pulled mid-call from the documents you uploaded.

  3. 03

    Books the next step

    Schedules the callback and scores the lead while the caller is still on the line.

  4. 04

    Writes the report

    Summary, answers and outcome land in the dashboard before the next number dials.

Replay · outbound · screening
EngineRealtime s2s
Langta-IN · Tanglish
First audio579 ms
Barge-inBackchannel-aware

Illustrative replay. Tool names and first-audio times are from the production build.

02Sound familiar?

If this sounds like your week, I can help.

Most of my projects start with one of these. Each has a system behind it that's already running for a client.

01

Leads go cold because nobody calls them back in time.

A voice agent calls every lead within minutes, in their language, answers questions and books the next step.

Case study Zillion Connects
02

Your team reads PDFs and re-types the same fields every day.

Document AI pulls out the fields, flags what's missing or expiring, and answers questions with the page it came from.

Case study Maritime Compliance Copilot
03

Reports are built by copy-pasting from five spreadsheets.

An automation pulls the data, reconciles it and has the report ready before anyone asks.

Case study StockPulse
04

You tried a chatbot. It made things up.

Grounded answers with citations, an evaluation set that proves accuracy, and a no-AI path where mistakes aren't allowed.

Case study Lawyer Friend

03Results

Measured, not promised.

From systems running for real clients, on real client data.

300–500 msreply latency on live phone calls, down from 2–3 s
11Indian languages on live calls
−36%cost per minute after model tuning
19,136SKUs tracked from plain stock reports
27 / 27tables verified before a live cutover

04Selected work

Six builds, with the numbers.

Each one is written up the way I'd explain it to your team: what was broken, what I built, and what I measured.

01Voice AI2026Live

Zillion Connects

A multi-tenant SaaS that runs outbound AI phone campaigns. The agent speaks natural Tanglish or Hinglish, answers from a knowledge base, books callbacks, and has the call report written before the next number dials.

  • 300–500 msreply latency, down from 2–3 s on the first Twilio + Whisper pipeline
  • 11 languagesTamil, Hindi, Telugu, Malayalam, Kannada and six more, code-mixing included
  • 27 / 27tables verified row for row in the live AWS → VPS migration
FastAPIOpenAI RealtimeGemini LiveVobiz
Case study ↗

02LLM & RAG2026Pre-launch

Lawyer Friend

Legal-tech product, Tamil Nadu

Answers legal questions in Tamil or English strictly from a curated knowledge base, and produces the whole chain of Right to Information filings: the request, both appeals and the complaint.

  • 5 filingsRTI 6(1), two 19(1) variants, 19(3) and 18(1), in Tamil and English
  • 100%retrieval recall@6 and citation precision on the draft eval set
  • 4.86 / 5Tamil fluency, LLM-judged, before expert review
FastAPIPostgreSQLOpenAItext-embedding-3-small
Case study ↗

03LLM & RAG2026Pilot

Maritime Compliance Copilot

DataReveal AI

Reads registration certificates, inspection reports and MARPOL / SOLAS circulars. It extracts the fields that matter, raises risk flags, and answers fleet-wide questions with the file and page each answer came from.

  • 6 flagsexpired, expiring in 30 or 90 days, missing fields, repeat incidents, detention keywords
  • File + pagecited on every answer; the model may only use the documents
  • No-key moderetrieval hits and rule-based insight still work without an LLM
FastAPIPyMuPDFspaCyChromaDB
Case study ↗

04Automation2026Client

Jewellery Photo Pipeline

Jewellery vendor

An n8n pipeline that watches a Drive folder and enhances every product photo two ways with Gemini. For pieces where a single altered stone is unacceptable, a deterministic path runs with no AI at all.

  • 2 AI variantsa detail-safe retouch and a white-background studio shot, generated in parallel
  • 2.4×edge detail on the no-AI GraphicsMagick path, at zero API cost
  • Safe deleteoriginals leave the inbox only after every variant has uploaded
n8nGemini 2.5 Flash ImageGraphicsMagickGoogle Drive API
Case study ↗

05Automation2026Pilot

StockPulse

Electronics distributor

Turns a retailer's periodic stock exports into movement, velocity and reorder signals by diffing consecutive snapshots. No sales history needed.

  • 19,136 SKUsfrom the client's real exports, 3,782 of them in stock
  • ₹5.16 Crof stock valued at selling price and ranked by priority
  • Exact diffsmovement arithmetic checked for exact equality by a test suite
FastAPIpandasopenpyxlxlrd
Case study ↗

06Product2026Pre-launch

G2M

GOFTUS

A multi-tenant platform for gym owners covering members, coaches, attendance, class packs, payments and reminders. It runs on web and mobile and is sold on monthly plans.

  • 92 / 92backend tests passing after a full security-hardening pass
  • 3 clientsReact web app, Flutter mobile app and marketing site on one API
  • ~54 endpointsacross 11 routers, every query scoped to its gym
FastAPISQLAlchemy (async)AlembicReact
Case study ↗

Have something like this in mind? Tell me about it. The first call is free.

Book a free call↗

05Services

What I can build for you.

Most projects mix two or three of these. A voice agent is also a knowledge base, a CRM hook and a dashboard.

Outbound follow-ups, screening, reminders and inbound reception on real phone lines. They speak Tamil, Hindi and English the way your customers actually do, answer from your knowledge base, book the next step and log everything.

See it in production: Zillion Connects

Good for

  • Lead follow-up & qualification
  • Recruitment screening
  • Appointment booking & reminders
  • After-hours reception

You get

  • An agent on your phone number
  • Live call console & transcripts
  • Post-call summaries into your CRM
  • A cost-per-minute report

Contracts, certificates, invoices, SOPs. Pull out the fields that matter, flag what's missing or about to expire, and ask questions in plain language, with the file and page every answer came from.

See it in production: Maritime Compliance Copilot

Good for

  • Compliance & expiry tracking
  • Invoice & form extraction
  • Internal knowledge assistants
  • Drafting from templates

You get

  • Ingestion + OCR pipeline
  • Search with citations
  • Dashboards & alerts
  • An evaluation set that proves accuracy

WhatsApp flows, lead routing, daily reports, reminders, image and file pipelines. Built on n8n or plain Python workers and wired into the tools you already pay for.

See it in production: Jewellery Photo Pipeline

Good for

  • WhatsApp assistants
  • Lead routing into CRMs
  • Scheduled reports & reminders
  • Photo & document pipelines

You get

  • Self-hosted n8n or workers
  • Error handling & retries
  • Run logs you can read
  • A handover walkthrough

Multi-tenant SaaS, admin consoles, web and Flutter apps, payments and auth. Then deployment, migrations and hardening as it grows.

See it in production: G2M

Good for

  • MVPs for founders
  • Internal tools
  • SaaS with subscriptions
  • Web + mobile apps

You get

  • A production codebase with tests
  • Payments (Razorpay, Dodo)
  • Deployment on your cloud
  • Runbook & docs

06Process

From first call to production in weeks.

Short loops, your real data early, and no big reveal at the end. You see it working every week.

  1. Day 1

    Free scoping call

    30 minutes on the process you want gone. You get a one-page scope back: what 'done' means, the risks, and what it will cost to build and to run each month.

  2. Week 1

    Working prototype

    A number you can call, a document you can query, a workflow that runs on last week's files. Your real data, before the big spend.

  3. Weeks 2–6

    Build in slices

    One working slice at a time, a demo every week and a staging link that's always live. Latency, accuracy and cost are tracked from day one.

  4. Launch

    Ship & hand over

    Deployed on your accounts with a runbook, tests and a walkthrough for your team. Ongoing support on a monthly retainer if you want it.

Ways to work together

Fixed-scope build

A defined system with a fixed quote. Best when the process is clear and you want it done.

Monthly retainer

Ongoing AI and automation work, without hiring a full-time AI team.

White-label partner

For agencies and software firms: I build under your brand and stay behind the scenes.

07Why me

Why teams bring me in.

End to end

One engineer from the phone line to the database.

No hand-offs between five vendors, and no finger-pointing when something breaks at 2 a.m.

In production

Zillion Connects● Live
Plant feed monitoring● Live

Running for real clients, not in a demo.

Languages

Tamil

Supported: Tamil, Hindi, Telugu, Malayalam, Kannada, Marathi, Bengali, Gujarati, Punjabi, Odia, English

Speaks your customers’ language.

Measured

Before
2–3 s per reply · STT → LLM → TTS
After
300–500 ms · realtime speech-to-speech

Measured in milliseconds and rupees.

Every build ships with the numbers that matter: latency, accuracy, and cost per call or document.

Speed

1week

To a working prototype on your own data.

A number you can call or a document you can query, before the big spend.

Partners

7+companies

White-label friendly.

Agencies and software firms ship my builds under their own brand.

08Index

Everything else, in one list.

Client builds, pilots, reference architectures and a few tools I made for myself. Filter by the kind of work you need.

YearProjectKindBuilt withStatus
2026Live AWS → VPS migrationScripted cutover of a production voice platform: checksummed backups, a restore test, a freeze window, 27/27 tables verified.InfraBash · Docker · PostgreSQL · CaddyLive
2026Plant feed monitoringHourly feed and fuel tracking for a fertilizer plant. Tamil + English UI, alerts for missed entries and low stock, full audit log.ProductFastAPI · PostgreSQL · React · Vercel · RailwayLive
2026Tamil voice agent pilotFirst phone agent for Zillion, on Twilio then Vobiz. 266 ms to first audio in English, 579 ms in Tamil, with a per-minute cost report.Voice AIFastAPI · OpenAI Realtime · ElevenLabs · Sarvam AIClient
2026Textile payment collectionInvoices read by OCR, payments allocated FIFO, statements with a UPI QR, and reminders over WhatsApp and email on a schedule.AutomationFastAPI · Celery · Tesseract · WhatsApp API · ReactClient
2026Claude-driven video editorTalking-head footage to reels via plain-language edits: WhisperX cuts, animated captions, ducked audio, −14 LUFS export. Tamil supported.AutomationWhisperX · FFmpeg · Hyperframes · PythonPrototype
2026LeadFlow CRMMeta Lead Ads webhook ingestion, a Kanban pipeline, a unified inbox and a rule engine with 4 triggers and 6 actions.AutomationNext.js · Prisma · PostgreSQLPrototype
2026DocQueryDocument Q&A with no LLM. Hybrid FAISS + BM25 + fuzzy retrieval with MMR re-ranking; answers come back as steps and fields.LLM & RAGFlask · FAISS · BM25 · scikit-learnPrototype
2026Graph-enhanced fraud detectionA sliding-window transaction graph with 15 features and fraud-ring detection. AUC rises from ~0.76 to 0.90+ once graph features are added (synthetic data).Data sciencePySpark · Kafka · NetworkX · XGBoost · SHAPReference

09FAQ

Questions clients ask.

Anything not covered here? Ask me directly. The first call is free and there’s no pitch.

Book a free call↗

Every build gets a fixed quote after the free scoping call, so you know the number before any work starts. Running costs (models, telephony, hosting) are estimated up front too, per month and per call or document.

In the first week: a prototype on your own data. Most first versions go live in 2–6 weeks, depending on how many systems it has to connect to.

Yes. Everything is deployed on your accounts, and the code, prompts and data are yours.

Yes. Several partner companies white-label my builds for their own clients, and I stay behind the scenes.

Tamil, Hindi, Telugu, Malayalam, Kannada, Marathi, Bengali, Gujarati, Punjabi, Odia and English, including code-mixed speech like Tanglish and Hinglish.

Then I'll tell you. Some of my most reliable pipelines have a no-AI path for the steps where accuracy isn't negotiable.

Yes, fully remote. IST overlaps comfortably with the Gulf, the UK and Europe.

Hajik RenaaAI & Automation Engineer
Chennai, India

10About

I’m Hajik. I build the AI part and the unglamorous parts around it, because that’s where most AI projects actually fail.

The telephony, the queues, the database migrations, the admin screens. Clients bring me a process that’s eating hours (calls nobody has time to make, PDFs nobody wants to read, spreadsheets copied by hand every morning) and I turn it into something that runs on its own and can be measured.

I work with Claude Code and a spec-first workflow: written requirements, generated plans, then building and testing in small slices. It’s how one engineer ships what used to need a small team.

Based in
Chennai, India · remote
Speaks
Tamil, English
Works with
Founders, agencies, operations teams
Best fit
A painful process and real data to test on