Md Balal Raza
engineering • product • AI
I build highly predictable, cost-optimized AI production systems.
Product-Minded Software Engineer specialized in AI integrations and high-performance backend pipelines. I design resilient, multi-model architectures that solve real-world operational bottlenecks and scale with strict unit economics.
Client-Side Semantic Search
Runs a 22MB MiniLM language model completely in your browser via Transformers.js. Your search query is processed by a tokenizer and embedded locally to compute real-time semantic similarity against pre-calculated vectors. Zero data leaves your device.
Production Case Studies
AI Math Tutor: Guided Doubt Solving
To help students master concepts themselves, we engineered a spatial AI Math Tutor utilizing Gemini 3.1 Pro and Flash. The interface offers interactive question bank selection or physical worksheet image upload (OCR). While students work, the AI provides real-time guidance and highlights mistakes visually by drawing red circles directly around incorrect parts of equations, mimicking a human tutor.
Enterprise AI Voice Automation Pipeline
High-volume Ed-Tech demo reminders and rescheduling created an expensive human operational bottleneck at Bhanzu. This Event-driven WebSocket system automates the workflow deterministically using live scheduling infrastructure.
Work Experience
- AI Smart Tutor (Feb 2026 - Apr 2026): Developed a spatial UI-driven AI tutor that facilitates step-by-step mathematical problem solving. Built a dynamic scaffolding engine that provides personalized hints, has real-time error correction, and suggests missing steps to students. [Read the Full Engineering Case Study ↑]
- AI Voice Automation Pipeline (Sep 2025 - Oct 2025): Architected and deployed the core event-driven AI Voice Agent system for mid-funnel operations, driving an immediate 30% operational headcount reduction. [Read the Full Engineering Case Study ↑]
- Led complete rewrite of student dashboard using bulletproof-react architecture, building extensible infrastructure with comprehensive test coverage and development guardrails for improved team velocity.
- Built real-time classroom platform using AWS IVS serving 5,000+ concurrent users, delivering features like screen sharing, whiteboard, chats, picture-in-picture mode, quizzes, breakout rooms, in a 6-engineer team.
- Hackathon Winner: Designed and deployed an AI-based Stories and Highlights feature that automates video extraction from class recordings, streamlining student recaps, enhancing dashboard engagement. [Read the Medium Post ↑]
- Optimized company website from 56 to ~95 Lighthouse score on desktop and from 45 to ~85 on mobile.
- Led a 5-member team to develop and deliver the Streams platform within tight deadline.
Full Stack Developer Consultant
American Express
Remote- Legacy System Decoupling: Led the architectural migration of 3 monolithic enterprise applications, decoupling heavy server-side rendering (JSP) into a modernized, state-driven component architecture.
- Latency & Query Optimization: Engineered comprehensive database query optimizations and asynchronous data pipelines, slashing end-to-end application latency by 68% (3.8s to 1.2s) and significantly improving system throughput.
- Identity & Security Integration: Integrated Okta SAML2 SSO across legacy infrastructure, resolving critical security bottlenecks and reducing user onboarding friction by automating a multi-stage manual verification pipeline.
- Infrastructure Standardization: Architected standardized technical workflows and documentation for 15+ complex enterprise systems, cutting cross-functional developer onboarding time by 66%.
Full Stack Developer
BDI Plus Pvt. Ltd.
Bangalore, India- ETL Interface Architecture: Directed a 5-engineer team to build the core architecture for 'OneData', a complex ETL orchestration platform featuring real-time workflow state visualization, job scheduling, and dynamic data transformation pipelines.
- Automated Underwriting Systems: Deployed the core application layer for the 'OneDigital' insurance platform serving major clients (SBLI, HavenLife). Engineered automated pipelines handling 1,000+ complex policy calculations and real-time quote generations monthly.
- Augmented Analytics Rendering: Designed the client-side architecture for an Augmented Analytics (AugAn) engine, optimizing browser-side processing to handle massive data visualizations without main-thread blocking.
Founding Engineer
Finoramic Inc.
Hyderabad, India- 0-to-1 Fintech Architecture: Served as founding engineer, architecting the complete technology stack from ground zero to a production-scale platform spanning B2C personal finance and B2B lending markets.
- Algorithmic Data Pipelines: Engineered the core backend infrastructure, building a deterministic transaction deduplication algorithm and intelligent categorization engine that autonomously processed thousands of multi-source financial data points daily.
- B2B Credit Assessment: Built a high-performance B2B lending portal featuring real-time financial health scoring and algorithmic credit assessment tools for institutional lending partners.
- Operations Automation: Developed an automated, programmatic growth engine, orchestrating zero-code workflows that slashed manual operational marketing overhead by 80%.
Latest Articles
Operating Principles
Systems Over Snippets.
The current "AI wrapper" era is obsessed with the prompt; I am obsessed with the pipeline. Generative AI is inherently non-deterministic, making it highly volatile for enterprise systems. I build for the edge cases. My approach focuses on rigorous guardrails, evaluation metrics, and error-handling hierarchies. I design systems that assume the AI will be wrong—and fail gracefully when it is. I am not interested in building toys that work 80% of the time; I am interested in building resilient, cost-optimized engines that form the backbone of a scalable product.
Have an architectural bottleneck or a high-overhead workflow?
Let’s discuss how to optimize your business operations through deterministic AI pipelines and strict unit economics.
Technical Consultation
Edge AI Intent Filter (B.S. Filter)
Runs a lightweight MobileBERT language model completely in your browser via Transformers.js. As you type a message, it detects intent in real-time, labeling it as high-value, outreach, or spam. Zero data leaves your device.