Silicon Colleague is a production-grade, multi-tenant AI Security & Compliance Gateway purpose-built for Indian enterprises under the DPDP Act 2023. The system intercepts every LLM interaction in real time through a 10-stage compliance pipeline — detecting, masking, risk-scoring, and governing sensitive personal data before it reaches any foreign AI server. Built entirely solo with no funding, no team, no external support.

• Founded & independently architected from zero: Identified a gap — no Indian tool addressed the intersection of generative AI adoption and DPDP Act 2023 enforcement. Designed entire system architecture, database schema (10 tables, multi-tenant), API layer, NLP pipeline, and deployment infrastructure solo.
• Built a hybrid 3-layer PII detection engine with 1,500+ lines of detection logic: Combined regex, BERT-based NER (IndicBERT), and a custom 5,000+ entry Indian name dictionary spanning Hindu, Muslim, Sikh, Christian, Jain, and Parsi communities. 51 rules across 7 categories mapped to 21 Indian laws.
• Engineered a 4-tier automated risk-scoring engine: Continuous 0–100 scoring with co-occurrence multipliers, contextual modifiers, and Redis-backed repeat-offender escalation. Four legally defensible tiers: LOW → MEDIUM → HIGH → CRITICAL (hard block before any AI call).
• Identified and rebuilt a critical fail-secure architecture: Red-team testing revealed a silent CRITICAL→LOW risk downgrade on provider timeout. Rebuilt entire failure path: AI-provider failover (Gemini → Claude → human review), circuit breaker, exponential backoff. System never shows ALLOWED when the request actually failed.
• Self-learning review loop: 97.1% → 99.4% accuracy: Human × Agent Review Panel updates confidence model on every decision. Never stores raw PII — only masked patterns and surrounding context.
• Enterprise security architecture: bcrypt, httpOnly JWT, 3-tier RBAC, per-table multi-tenant isolation, tiered rate limiting, bot protection, Docker network isolation, HMAC-signed audit logs.
• Demonstrated to the Managing Partner of a $500M investment fund as a billable enterprise SaaS for Banking & Finance, Healthcare, Legal, E-Commerce, and Government sectors.

opsell.ai is an AI-powered sales intelligence platform. Working on the backend engineering team, directly responsible for the data infrastructure and normalization systems powering the platform's AI/ML workflows — ensuring data quality, consistency, and reliability across the entire stack.

• Engineered a scalable multi-source data normalization engine: Python + FastAPI pipeline that ingests, cleans, standardizes, and structures data from multiple heterogeneous sources with differing schemas. Applies rule-based validation, type coercion, field normalization, and deduplication logic — ensuring downstream AI/ML models receive clean, consistent training and inference data.
• Designed and maintained core REST API backend architecture: Developed and maintained RESTful API endpoints on PostgreSQL with input validation, structured error handling, response schema consistency, and query optimization. Collaborated cross-functionally with ML and product teams.
• Improved data pipeline reliability: Identified and resolved data quality issues causing inconsistent behaviour in downstream ML models. Implemented validation layers and monitoring checkpoints at key pipeline stages — reducing silent data corruption failures.
• Contributed to backend architecture decisions: Data schema design, API versioning strategy, and pipeline scalability — bringing production-architecture thinking from Silicon Colleague into a commercial team environment.