Process Automation & AI Developer
I build cost-effective, self-hosted AI systems and automated pipelines that solve complex operational challenges — bridging 10+ years of finance operations with modern AI engineering to eliminate manual bottlenecks.
I don't just write code — I deeply understand the workflow frictions and legacy systems that slow teams down, and bridge that gap with modern AI.
I build cost-effective, self-hosted AI systems and automated pipelines designed to solve complex operational challenges.
With 10+ years of experience in finance operations, I've lived through the manual bottlenecks that drain teams: month-end reconciliations, invoice sorting, ledger consolidation. Now I eliminate them with local LLMs, multi-agent orchestration, and enterprise data engineering.
Chennai, Tamil Nadu · Remote-ready
A focused toolkit spanning AI orchestration, automation, enterprise data, and infrastructure.
Autonomous systems, zero-cost local AI, and enterprise data pipelines — each solving a real operational problem.
Fully autonomous multi-agent network: an OpenClaw router triages prompts from a Telegram frontend and delegates to specialized Hermes executors for reporting, database lookups, and schedule notifications — all on a single VPS.
Wasted significant daily hours manually compiling reports, monitoring server states, and resolving scheduled execution logs.
Design a secure, zero-friction system triggering background jobs and query retrievals directly from a mobile messaging client.
Configured an OpenClaw gateway connected to Telegram, with a router agent parsing messages to select between research, execution, or notification pipelines powered by Hermes model endpoints.
Reclaimed ~70% of routine task time with 24/7 autonomous task dispatching on an inexpensive 8GB RAM VPS.
Custom self-hosted LLM environments using Ollama, llama.cpp, OpenWebUI, and AnythingLLM — with optimized context windows and quantization delivering sub-4-second responses on standard CPU with zero ongoing API costs.
High cloud API bills and data privacy concerns with proprietary remote model inference.
Run multi-gigabyte models locally with under-4s latency for conversational chatbots on consumer-grade hardware.
Used GGUF quantization via llama.cpp and Ollama, tuning memory mapping, context buffers, and thread assignments to maximize CPU inference velocity.
Maintained a zero-cost local stack with sub-4s responses, saving an estimated ₹8k/month in cloud API equivalence.
End-to-end Tally-XML-to-Excel integration automating month-end accounting reviews. Advanced Power Query (M code) ingests, cleans, and consolidates ledger XML files into auto-reconciled financial reports and live dashboards.
Month-end consolidation and MIS generation for six travel operations entities took two full working days and was prone to copy-paste ledger discrepancies.
Automate data ingestion, schema formatting, and bank/ledger reconciliations in Microsoft Excel.
Wrote an end-to-end M Code script to parse Tally XML exports, filter ledger codes, structure hotel/visa datasets, and output formatted pivot-ready tables.
Cut closing time from 2 days to 15 minutes with 100% mathematical accuracy and zero manual editing.
Python suite that extracts GST-compliant data from airline invoice PDFs (Air India, Indigo, SpiceJet, Air India Express), reconciles them against Excel ledgers with OCR fallback for scanned files, and splits/sorts consolidated PDFs by company — replacing hours of manual sorting and re-keying.
Hundreds of airline GST invoice PDFs (Air India, Indigo, SpiceJet…) had to be sorted by airline, matched to Excel voucher ledgers, and re-keyed into the ERP every month — consuming hours and inviting errors.
Automate extraction, matching, and filing so the accounts team gets clean, ERP-ready data without opening a single PDF.
Built a Python suite: per-airline extractors parse invoice metadata and CGST/SGST/IGST totals via pdfplumber text + table extraction; a reconciliation engine auto-detects Excel headers, matches voucher numbers to PDFs using exact and normalized variants, and runs OCR on scanned files; a slicer splits consolidated PDFs into per-company files from a Voucher → Ledger map.
Invoice data lands as structured Excel in minutes with exception flags and confidence levels, eliminating manual sorting and re-keying.
High-performance, responsive web interfaces featuring glassmorphic design, dynamic data bindings, and interactive simulators — optimized for speed, accessibility, and clean visual storytelling.
An AI assistant embedded in this portfolio, wired to any OpenAI-compatible LLM (OpenAI, DeepSeek, Ollama, OpenWebUI). The API key stays in build-time env vars, with a static knowledge-base fallback so the widget stays useful even without a configured model.
Messages arrive from your phone, are authenticated, and handed to the router.
From finance operations to autonomous AI systems — a decade of turning operational friction into automated workflows.
Architecting self-hosted AI solutions, multi-agent frameworks, and local LLM installations. Deploying models on optimized VPS environments, building autonomous Telegram workflow bots, and bridging legacy business software with modern LLM capabilities.
Managed credit operations, payment reconciliations, and financial reporting. Spearheaded internal process automation, converting multi-day manual audit checks into rapid scriptable workflows that minimized human error and saved hundreds of processing hours.
Directed end-to-end accounting processes, vendor settlements, and billing workflows. Designed automated Excel and Power Query templates that optimized revenue auditing and reporting efficiency for travel operations.
"Balaji took a manual Excel reconciliation process that took our teams two days every month and automated it completely using Power Query. It now runs in 15 minutes with zero mistakes."
"He has a unique combination of deep accounting background and technical AI automation. He built a multi-agent system on our VPS that processes support requests flawlessly."
Have an operational bottleneck or a process you'd love automated? Request a free 15-minute audit.
Prefer email or a quick call? Reach me directly — I usually reply within a day.