Generative AI Company in Chicago

Regulated logging is the line item that separates a Chicago GenAI build from a generic one: a scoped Chicago generative AI proof of concept over six to ten weeks, covering data audit, a RAG baseline or fine-tuned model, an evaluation harness, BIPA scoping if applicable, and a hosted demo with logging, a custom production copilot (CME or Citadel research analyst copilot, Allstate claims summariser, McDonald's drive-thru intent classifier, Walgreens pharmacy support deflection) including evaluation, prompt injection hardening, MRM or insurance regulator documentation, BIPA consent infrastructure if applicable, and SEC 17a-4 or CFTC 1.31 immutable logging where relevant, or a full production generative AI system with multi-step agents, multiple models, fine-tuning, and enterprise integrations. USD, fixed-fee.

BIPA (740 ILCS 14) defines biometric identifier as a retina or iris scan, fingerprint, voiceprint, or scan of hand or face geometry, and biometric information as any information based on a biometric identifier used to identify an individual. The Cothron v. White Castle (2023) Illinois Supreme Court ruling held that a separate claim accrues with each scan, which dramatically increases statutory damages exposure for any system that repeatedly processes biometric data. For generative AI this matters whenever the system ingests voice (voiceprint formation), images of faces (face geometry), or fingerprint or iris data, including indirect cases like a copilot that summarises call recordings or a vision model that captions retail security footage. Our designs include explicit written consent capture before first biometric processing, a published retention and destruction schedule (typically three years from last interaction or one year after initial purpose is fulfilled, whichever comes first), no sale or profit from biometric data, encryption in transit and at rest, and on-device processing where possible so the raw biometric never transits the wire.

Yes. We ship investment and trading research copilots, regulatory filing automation (Form CPO-PQR, Form PF, Form ADV, 10-K, 10-Q), trade reconstruction summarisation, market commentary drafting, and code copilots for quant research at CME Group, Cboe Global Markets, DRW Holdings, Citadel Chicago, Jump Trading, GTS, Optiver Chicago, and Belvedere Trading. Deployments stay inside SEC Rule 17a-4 immutable retention (typically AWS Bedrock with CloudTrail to an S3 bucket under Object Lock in Compliance mode, or the firm's existing 17a-4 archive on Mimecast, Smarsh, or Global Relay), CFTC Rule 1.31 commodity recordkeeping (5 years, last 2 readily accessible), NFA 2-10, FINRA 4511, and the firm's MRM framework modelled on SR 11-7. Every prompt, retrieval, and completion is logged for audit. We coordinate with the firm's compliance and MRM teams; we do not pretend to replace them.

The Illinois Artificial Intelligence Video Interview Act (820 ILCS 42), effective 2020 and expanded 2022, requires employers using AI to analyze video interviews of Illinois applicants to notify the applicant before the interview, explain how the AI works and what general characteristics it evaluates, obtain consent, share the video only with people whose expertise is necessary to evaluate the applicant, destroy the video and copies within 30 days of an applicant's request, and (for employers relying solely on AI in deciding which applicants advance to in-person interviews) report demographic data annually to the Illinois Department of Commerce. Any generative AI hiring assistant for Illinois roles, including resume-screening copilots that read transcribed video, must build these surfaces in. We design them in by default and we will not ship a hiring tool for Illinois roles without them, regardless of who the client is.

Yes, with the right architecture. McDonald's was sued under BIPA in 2021 (Carpenter v. McDonald's) over drive-thru voice AI, and Walgreens has faced BIPA litigation over biometric employee timekeeping. Drive-thru voice ordering captures voiceprints, and any pharmacy or retail kiosk that does identity-by-face triggers BIPA face geometry obligations. We design with on-device voice processing (Hailo-8 or NVIDIA Jetson Orin Nano running speech-to-text locally so raw audio never leaves the endpoint), explicit signage at the order point disclosing voice capture, an opt-in consent flow tied to the loyalty app, and a retention schedule capped at trailing 90 days for unmatched audio. The generative AI layer (intent classification, order completion suggestion, upsell drafting) operates on the text transcript, not the raw voice, which materially reduces BIPA blast radius. We also map the design against the FTC's 2023 guidance on AI voice cloning.

When a project requires genuine research (novel architectures, interpretability work, alignment research, formal verification, rare-event detection in trading data) we scope collaborations with University of Chicago (including the Booth School and the Data Science Institute), Northwestern (Evanston, including the Kellogg School and the Computer Science department), UIUC (Champaign, the Coordinated Science Laboratory), Illinois Tech, Loyola, or DePaul faculty rather than overselling in-house capability. Our core team handles applied engineering, RAG, fine-tuning, eval harnesses, MLOps, and productionisation, which is where most Chicago generative AI projects actually stall. For standard work (RAG over enterprise knowledge, fine-tuning Llama or Mistral on client data, evaluation, prompt injection hardening) no academic partner is needed. We will tell you up front which bucket your problem fits into.

A research analyst copilot at CME Group, Cboe, Citadel Chicago, DRW, or Jump Trading scale takes fourteen to twenty-four weeks from kickoff to production, including SEC 17a-4 and CFTC 1.31 immutable retention, MRM documentation, and prompt injection red-teaming. An insurance claims summariser or underwriting copilot at Allstate, State Farm, or CNA scale takes sixteen to twenty-eight weeks including NAIC Model Bulletin 9999 alignment and Illinois Department of Insurance review. A QSR or retail support copilot at McDonald's, Walgreens, or Ulta scale takes twelve to twenty weeks including BIPA consent infrastructure where biometric inference is in scope. A biopharma protocol or regulatory copilot at AbbVie, Baxter, or Takeda US scale takes fourteen to twenty-four weeks including 21 CFR Part 11 audit trails. Fixed-fee.

Illinois charges a 4.95% flat individual income tax and a 7% corporate income tax, plus a 2.5% personal property replacement tax on corporations (effectively 9.5% combined). Chicago imposes no city income tax but does levy a 9% lease and rental tax (Chicago Personal Property Lease Transaction Tax, Code 7550) that touches certain SaaS contracts, including a recent technical bulletin clarifying generative AI services delivered as SaaS. Illinois sales tax is 6.25% state plus local (10.25% in Chicago), and Illinois does not generally tax services, so design and engineering fees ship without sales tax. The Chicago lease tax can apply to nonpossessory computer leases (cloud services and certain AI APIs) and may need to be invoiced separately depending on contract structure. We invoice in USD, accept ACH and wire, provide W-9 documentation, and route international clients through Canadian or Indian entities under treaty.