Applied AI & LLM engineering

We take language models from demo to dependable service. That means schema-enforced APIs, retrieval that is grounded in your data, agents with clear tool boundaries, and honest evaluation against simple baselines before anything ships.

• RAG and retrieval pipelines over proprietary data
• Structured, schema-validated model outputs
• Agent and tool-use design with guardrails
• LoRA fine-tuning of open-weights models
• Local inference with Ollama on commodity hardware
• Replayable evaluation harnesses with baselines
• Token and cost reduction: caching, dedup, routing

Trading & fintech systems

We have built and operate a production trading platform end to end: signal ingestion, order routing across multiple brokers, verified execution, ledger reconciliation and per-strategy risk caps. We bring that operating experience to client systems.

• Broker integrations: Interactive Brokers, Tradovate, Alpaca
• Futures, options (including SPX 0DTE) and equities
• Verified execution and broker-to-ledger reconciliation
• Risk guardrails: daily caps, give-back protection, kill switches
• Strategy research, backtesting and shadow-mode validation
• TradingView and Pine Script tooling

Data platforms & pipelines

Reliable data is the boring half of every AI project. We build ingestion with retries and dedup, time-series stores with retention policies, replay tooling for backtests, and dashboards that show what the system actually did.

• Market and event data ingestion
• MongoDB, PostgreSQL, Prisma, Parquet
• Backtest and replay infrastructure
• Operational dashboards and audit trails

Product engineering

Small senior teams shipping complete products: TypeScript front to back, typed APIs, CI on every pull request, and deployment on the cloud provider that fits your budget.

• Next.js, React, TypeScript, Tailwind
• Node.js and Python (FastAPI) services
• CI/CD with GitHub Actions, Docker
• AWS, Vercel, Render and similar

Speech & Indic-language AI

We evaluate and integrate speech models for Indian languages, where general-purpose tools often fall short. Batch pipelines, vendor comparisons and quality checks that hold up on real audio.

• Batch transcription pipelines
• Vendor evaluation (Sarvam, Whisper and others)
• Post-processing and quality review workflows

AI adoption & automation

Most teams do not need a research project; they need one workflow automated well. We find it, build it, measure it, and teach your team to own it.

• Workflow and process automation
• Prompt design with measured outcomes
• Internal tools and copilots
• Team enablement and code review